Robert Cheska was charged with mail fraud in conjunction with a scheme to kill horses for insurance money. He was convicted by a jury, but the district court granted a new trial because of a remark made by the prosecutor during closing arguments. Because of the broad discretion accorded district court judges in deciding whether to grant a new trial, we affirm.
I.
Cheska and his co-defendant, George Nuber, were charged with using the United States mails to defraud an insurance company in connection with the death of a show horse owned by Nuber. Over the years, Nuber purchased a number of horses from Cheska, a professional horse trainer. Nuber boarded his horses at stables operated by Cheska, and Cheska trained Nuber’s daughters and their horses in equestrian activities. Nuber purchased a horse for one of his daughters in the spring of 1986, using Cheska as the agent for the transaction. Nuber paid $6500 for the horse, which was originally named Wanja, then renamed Jolly Roger, and finally renamed Valentino. Apparently, the horse did not work well with Nuber’s daughter, and she wanted another horse named Silver Rabbit, which was owned by Cheska’s father. Nuber decided to sell Valentino in order to buy Silver Rabbit.
What happened next is in dispute, and the jury ultimately acquitted Nuber on mail fraud charges, so we must assume they were not convinced by the government’s version of events as they related to Nuber. In any event, the evidence showed that Nuber had insured Valentino’s life for its full value, up to $50,000. In January 1987, Nuber and Cheska attended a horse show in Wisconsin, where Cheska met with Timothy Ray a/k/a Tommy Burns, a longtime friend. Cheska told Burns that Nu-ber wanted to have Valentino killed and was willing to pay $5000. Burns, who had killed several other horses for insurance money, readily agreed. The two planned for Valentino to be moved to Florida, where Burns was living, so that the killing could be accomplished. After the horse arrived in Florida, Burns electrocuted it in order to make the death appear to be from natural causes'. Burns’ girlfriend, Lisa Kinney, acted as his lookout.
*949Nuber submitted a claim to his insurer, including a sworn proof of loss valuing Valentino at $50,000, the policy limit. Nu-ber claimed he had paid for Valentino with cash and trade-ins worth $50,000, and attempted to establish the value with affidavits and bills of sale. The insurer, apparently skeptical of the value claimed and the circumstances surrounding the death, hired a veterinarian to perform an autopsy. The veterinarian could not determine the cause of death, but found the evidence was consistent with death by electrocution or being struck by lightning. The insurer ultimately refused to pay on the claim, and Nuber sued the insurer in state court. In the course of this lawsuit, Nuber and Cheska told a number of conflicting tales about the value of Valentino, how Valentino was purchased, and what was paid in cash and trade for the horse. Not surprisingly, Nuber lost his case before a jury, and the federal government took an interest in the allegedly false documents Nuber and Cheska created and transmitted through the mails in their attempts to establish the value of Valentino. The government charged Nuber and Cheska with mail fraud. A jury convicted Cheska but acquitted Nuber.
At trial, Tommy Burns testified for the government, as he had in a number of trials relating to the killing of horses for insurance money. It was a topic with which Burns showed a disturbing familiarity. By the time of Cheska and Nuber’s trial, Burns had personally killed fourteen other horses at the request of their owners, all for the sake of insurance fraud. Burns struck a very favorable plea agreement, which required him to cooperate with the government in its investigation and prosecution of the crimes in which he took a pari. In exchange for his cooperation, Burns received a six month federal prison sentence, which he had already served in full at the time of Cheska and Nuber’s trial. Needless to say, Burns was subject to an intense attack on his credibility at Cheska and Nuber’s joint trial. In cross-examination and during closing arguments, both defense counsel called Burns’ credibility into question. Counsel for Cheska additionally argued that Lisa Kinney, Burns’ girlfriend and lookout, had lied during her testimony at trial as well. In rebuttal, the prosecutor sought to undo the damage with the following remarks:
Why would Lisa Kinney come into federal court and lie? To help Tommy Burns? Tommy Burns’ deal is over. Tommy Burns has served his sentence. Tommy Burns has nothing else to gain at this point. Tommy Burns has convicted 23 other people. What is his motivation at this point?
Tr. p. 2401. These remarks drew immediate objections from defense counsel and a motion for a mistrial. The court issued a curative instruction, and took the motion for a mistrial under advisement. After the jury convicted Cheska and acquitted Nu-ber, Cheska renewed his motion for a mistrial, and this time, the district court granted the motion and ordered a new trial.
The district court found that the remark that Burns had convicted 23 other people was literally untrue, and was based not on the evidence but on the prosecutor’s personal opinion. The court reasoned that the remark was therefore improper, and next considered whether the remark had the effect of denying Cheska a fair trial. Because Burns was the government’s key witness against Cheska, and because his credibility was at issue, the court was concerned about unfair prejudice to Cheska that could result from an improper bolstering of Burns’ credibility. The court found that its corrective instruction was inadequate to remedy the effect of the remark on the jury, especially in light of the fact that the evidence of Cheska’s guilt was “far from overwhelming.” Therefore, the court “reluctantly” ordered a new trial. The government appeals from that ruling.
II.
The district court has great discretion in deciding whether to grant a new trial based on prosecutorial misconduct. *950 United States v. Henry, 2 F.3d 792, 794 (7th Cir.1993). This is because the trial court judge is in the best position to determine the seriousness of any incidents that occurred during the trial. Id. We have outlined a methodology for the district court to use in making this determination. See Henry, 2 F.3d at 794; United States v. Butler, 71 F.3d 243, 254 (7th Cir.1995). First, the court should consider whether the remark was improper. If not, the inquiry is over, and there is no reason to grant a new trial. If so, the court should evaluate the remark in light of the entire trial to determine whether it deprived the defendant of an fair trial. Butler, 71 F.3d at 254. In assessing the effect of the remark on the fairness of the trial, the court should consider:
1) the nature and seriousness of the prosecutorial misconduct; 2) whether the prosecutor’s statements were invited by conduct of defense counsel; 3) whether the trial court instructions to the jury were adequate; 4) whether the defense was able to counter the improper arguments through rebuttal; and 5) the weight of the evidence against the defendant.
Butler, 71 F.3d at 254. The district court engaged in this analysis and concluded that, on balance, a new trial should be granted.
The government claims the remark was not improper in the first instance because it was based on a reasonable inference from the evidence. In the alternative, the government contends that even if the remark was technically improper, 1) the effect of the remark on the jury’s assessment of Burns’ credibility was slight; 2) the bolstering of Burns’ credibility was a legitimate use of the remark; 3) Burns’ testimony was not necessary for the conviction; 4) the curative instruction was adequate; and 5) the weight of other evidence against Cheska was substantial. In sum, the government disputes the district court’s exercise of discretion in finding that Cheska’s trial was rendered fundamentally unfair by the prosecutor’s remark. We are not here, of course, to reweigh the considerations of the district court, or to decide if we would have come out the other way. When reviewing the district court’s ruling for abuse of discretion, we reverse only if we have a “strong conviction of error.” United States v. McClinton, 135 F.3d 1178, 1186 (7th Cir.1998), cert. denied, 524 U.S. 921, 118 S.Ct. 2308, 141 L.Ed.2d 167 (1998); Ladien v. Astrachan, 128 F.3d 1051, 1056 (7th Cir.1997) (we are obligated to affirm a decision under the abuse of discretion standard unless no reasonable person could concur with the trial court’s assessment under the circumstances). With that standard in mind, we review the government’s arguments.
A.
We first consider whether the remark was improper. The district court found that the statement that “Tommy Burns has convicted 23 other people” was not based on a reasonable inference from the evidence and was instead the product of the prosecutor’s personal opinion. The district court also noted that, earlier in the trial, it had declined to allow the government to bolster Burns’ credibility with the evidence of the results of other criminal cases in which Burns had provided information. The court ruled that the unfair prejudice of this evidence outweighed its probative value, and the court believed that the prosecutor’s remark in closing arguments was an attempt to circumvent this ruling. The government disavows any improper motive in making the remark, and explains that the prosecutor was merely trying to demonstrate to the jury that Lisa Kinney had no motive to lie to protect Burns at the time of Cheska’s trial.
The government concedes that the' remark was literally untrue because Burns could not personally convict anyone, that task being exclusively the province of a jury or a court. The government contends that because no juror would have interpreted the remark literally, the fact that it was literally untrue could not have ren*951dered it improper. Urging us to find that the statement is supported by the record, the government directs us to the following evidence admitted at trial without objection from Cheska: 1) when asked at trial who Jerry Farmer was, Burns replied, “Just one of the 23 people that were convicted in the horse crimes.” Burns later testified that he cooperated against Farmer and that Farmer was one of the people convicted in the horse investigation; 2) Burns testified that he admitted to federal authorities that he had killed fifteen horses for owners who were committing insurance fraud; 3) Burns testified that he cooperated with the government extensively, meeting with an FBI agent close to 200 times, telling the government about a number of his friends,and discussing the cases with the government over a number of years; 4) Burns testified that his plea agreement required him to cooperate through the rest of the investigation, that his cooperation extended beyond Cheska’s trial, that he had “testified a lot,” and had testified before two other judges in earlier trials; 5) Burns testified that as a result of his cooperation, he received a reduced sentence of six months in prison, a sentence he had already served; and 6) Burns testified that when he was sentenced in a related Florida state case, federal authorities told the Florida court of Burns’ cooperation in the horse investigation. The government asks us to conclude that from this evidence, it was only a “tiny” inferential step to conclude that Burns was a key figure in the horse investigations and that his cooperation had led to the convictions of 23 people.
Even if we accept the government’s explanation of the purpose of the remark, and even though the remark was isolated, delivered in the course of responding to defense attacks on Kinney’s credibility and not intentionally improper, we must still consider whether there was an adequate basis in the record to support it. In its order, the district court focused only on Burns’ testimony that Farmer was one of 23 people convicted in the horse crimes, and we cannot tell from the record whether the district court considered any of the other evidence that the government now cites in support of the remark. The government concedes that even if the jury assumed the remark meant that Burns was the key player in the convictions of 23 other people, the remark was still an exaggeration. And this is just one possible interpretation of the remark.1
Other possible inferences that could be drawn from the evidence the government cites are 1) that there had been 23 convictions; 2) that Burns’ contribution was decisive in 23 cases, and that but for Burns’ cooperation, those individuals would not have been convicted; 3) 23 juries convicted defendants based on Burns’ testimony; 4) a jury or juries convicted 23 individuals on the basis of Burns’ testimony; 5) Burns’ testimony forced pleas in 23 cases; or 6) Burns’ credibility was decisively established in 23 other cases. Some of these inferences are factually incorrect; others have some basis in the record. In any case, the remark is vulnerable in three key respects: the number of convictions, the extent of Burns’ contribution and the effect of Burns’ contribution. Contrary to the dissent’s assertion, we conclude not that the remark is improper because the *952jury may have drawn an improper inference. Rather, the remark was improper because it cannot be supported by the record in these three key respects, and these weaknesses are amply demonstrated by the sample inferences we have drawn.
A giant leap, not simply a “tiny step” was required to conclude that Burns was the person largely responsible for the convictions of 23 other defendants. The government conceded at oral argument that if the prosecutor had argued that 23 other juries believed Burns and that this jury should therefore believe him as well, the remark would be much more difficult to defend, yet that is a very likely interpretation the jury could have placed on the remark. The government urges us to find the inference reasonable under United States v. Waldemer, 50 F.3d 1379, 1384-85 (7th Cir.1995), cert. denied, 515 U.S. 1151, 115 S.Ct. 2598, 132 L.Ed.2d 845 (1995). There we stated that “[w]hether the evidence bears logical and proximate connection to the point the prosecutor wishes to prove are perhaps the most obvious considerations in determining whether the inference is reasonable.” Id. The prosecutor purported to make the remarks in response to an attack on Kinney’s credibility. He was attempting to rehabilitate her after the defense suggested she was lying to assist Burns, who was her boyfriend. With that standard in mind, we examine the prosecutor’s statements. The fact that Burns no longer needed Kinney’s assistance because he had struck a deal with the government, and had already served his sentence is certainly logically and proximately connected to the goal of rehabilitating Kinney. But the statement that Burns had already convicted 23 other people is completely unrelated to Kinney’s credibility. Rather, it addresses Burns’ credibility. Our conclusion is bolstered by what the prosecutor said immediately after commenting that Tommy Burns had already convicted 23 other people: “What is his motivation at this point?” Tr. at 2401 (emphasis added). The prosecutor had thus shifted to a discussion of Burns’ credibility, which was also under attack.
Of course, there was nothing improper about attempting, in general, to rehabilitate Burns in closing arguments, given the defense attacks on his credibility. The question remains whether the government was allowed to use this particular argument in rehabilitating Burns. The district court was distressed because it had specifically disallowed the prosecutor from raising Burns’ extensive involvement in other prosecutions, reasoning that the unfair prejudice of this fact outweighed the probative value. The government counters with our ruling in United States v. Lindemann, 85 F.3d 1232, 1243 (7th Cir.1996), cert. denied, 519 U.S. 966, 117 S.Ct. 392, 136 L.Ed.2d 307 (1996), where we addressed the use of evidence of cooperation in other cases to counter an attack on a witness’ credibility. Amazingly, it was Burns’ credibility that was at issue in Lin-demann as well, another case involving the killing of horses for insurance money. In that case, the trial court admitted over objection evidence that Burns had successfully cooperated in other cases. Linde-mann challenged that admission as improper bolstering of Burns’ credibility, but we affirmed on the grounds that the evidence was relevant to the issue of bias, because Burns’ successful cooperation in other cases made less probable the assertion that Burns was lying in this particular case out of self-interest. The district court in Lindemann also clearly instructed the jury that they were to use that evidence only in assessing Burns’ credibility and not consider it as direct evidence of the defendant’s guilt. In other words, the jury was instructed that it could not use evidence of convictions or guilty pleas in other cases as evidence of the defendant’s guilt, but could use it only to assess Burns’ credibility. We thus upheld the district court’s discretionary decision to allow the evidence in light of the court’s assessment of the effect of that evidence on the jury in the context of that trial. Lindemann, 85 F.3d at 1243-44.
So the government is correct that the district court might not have abused its *953discretion had it allowed evidence and argument relating to Burns’ successful cooperation. But it does not follow that the court in Cheska’s case was obliged to allow this evidence and argument. As we discuss below, the district court was in the best position to weigh the effect of the argument on the jury in light of the totality of the evidence at that point in time. In light of all of the improper inferences the jury could have drawn from the remark, and the size of the leap from the evidence in the record to the prosecutor’s remark, we conclude that the district court did not eiT in finding that the remark was improper in the context of the trial as a whole.
B.
We must next consider whether the district couit abused its discretion in finding that the remark deprived Cheska of a fair trial, and that the curative instruction was inadequate to remedy the problem. Here the government argues that the district court erred in applying the five-part test we set out above. Specifically, the government claims that the court grossly overstated the nature and seriousness of the remark, that bolstering Burns’ credibility was not unfairly prejudicial, that the court erred in finding that Burns’ testimony was crucial for conviction, that the court’s curative instruction was adequate, and that the weight of the evidence was substantial. The government conceded that two of the five factors do not militate in favor of reversal. In particular, the remark was not invited by any improper conduct of defense counsel, and because the remark was made in rebuttal, the defense had no opportunity to respond.
The vast majority of the government’s argument is directed to asking us to reweigh the considerations addressed by the district court. This is not our function when we are reviewing for abuse of discretion, and we decline the government’s invitation to be a Monday morning quarterback with the district court’s discretionary weighing of the various factors. The district court was in a far better position than we are to understand the impact of these remarks on the jury, in the context of the whole trial. We have only the cold, black and white record before us. The district court was there to see whether the jury bristled or gasped or perhaps sat impassively when the prosecutor made this remark. The district court was able to assess the credibility of the witnesses and the strength of the evidence presented and determine the impact of this comment in context. This sort of situation is precisely why we use the abuse of discretion standard, and we will not reverse unless it is clear from the cold record that no reasonable person could have ruled as the district court did.
Addressing the government’s assertions, we cannot find that the district court overstated the nature and seriousness of the remark. In the context of the trial, the district court judge believed this remark could seriously prejudice the jury. He issued a curative instruction:
Members of the jury, you should make your determination based upon the evidence, not upon any argument by counsel. It’s your collective recollection of the evidence that you should use in making your determination in the case.
Tr. at 2401-02. On further consideration, he determined this instruction was inadequate. It certainly was not as complete as the instruction given in Lindemann, where the court directed the jury that they were to use the evidence of Burns’ successful participation in other prosecutions only in assessing Burns’ credibility and not consider it as direct evidence of the defendant’s guilt in that case. Lindemann, 85 F.3d at 1243-44. The dissent correctly points out that we normally presume a jury will follow the court’s curative instruction. Here, however, the very judge that issued the instruction believed, in the context of the trial as a whole, that the instruction was inadequate to remedy the harm. Although we might not have ruled the same way, we cannot say that the district court’s discretionary decision regarding the effective*954ness of the curative instruction was an abuse of discretion.
Although it is true that bolstering Burns’ credibility was not an improper goal of closing arguments, as we discussed above, the issue is whether it was proper to use this particular evidence in asking the jury to draw the conclusion that Burns was not lying. The district court had already ruled that the evidentiary value of Burns’ successful participation in other prosecutions was outweighed by the harm of unfair prejudice. The district court likely feared that the jury would be unduly influenced by the fact that twenty-three other juries had already found Burns credible. Thus, although bolstering Burns’ credibility was a proper goal, the district court did not abuse its discretion in finding that the government would have to find some other way to accomplish that goal.
That leaves one final argument, whether the court erred in finding that Burns’ testimony was crucial for conviction, and whether the weight of the evidence was substantial. The government contends that the jury could have convicted Cheska even if it completely disbelieved Burns. The jury could have convicted Cheska under one of two theories, according to the government. First, they could have convicted Cheska if they found that Cheska and Nuber hired Burns to kill Valentino, and then Cheska lied to the insurer about the cause of Valentino’s death. Second, the jury could have convicted Cheska if they found that he submitted false documents and made false and misleading claims to the insurer regarding the true cost and value of Valentino. Burns’ credibility was crucial only to the first theory, the government contends, although the government concedes that Burns’ credibility was relevant, but not crucial, to the second theory as well. The government urges us to find that the jury could have “plausibly” convicted Cheska without believing a single word of Burns’ testimony. That may well be true, but again, we are reviewing the district court’s decision for abuse of discretion, and the district court was in a much better position than we are to determine the strength of the evidence for the alternate theories, and to predict the effect on the jury of these remarks. Regardless of the government’s view of the “ample evidence to convict,” the district court believed this was a close case. Indeed, Cheska’s co-defendant was acquitted, and so the jury apparently disagreed with the government’s “ample evidence” as well, at least as it regarded the evidence against Nuber. We are not left with a “definite and firm conviction” that the district court has erred in making that determination. McClinton, 135 F.3d at 1186.
III.
For all of these reasons, we conclude that the district court did not abuse its discretion in ordering a new trial for Ches-ka.
AFFIRMED.