These suits in admiralty, based upon the Death on the High Seas Act, 46 U.S.C. § 761 et seq., arose out of the crash of an Alouette II helicopter in the Gulf of Mexico off Leeville, Louisiana on November 30, 1959, killing the pilot and two passengers. , Libelants, in their representative capacities, sought recovery from Sud-Aviation, Societe Nationale De Constructions Aeronautiques (SUD), the manufacturer of the helicopter, claiming that a defect in the construction of the helicopter caused the crash. After a seven-day trial without a jury, on the issue of liability alone, the district court found against SUD, holding that the aircraft had been negligently manufactured and that SUD had breached an implied warranty of fitness.
This appeal by SUD questions whether the district court properly allocated the burden of persuasion and whether there was sufficient evidence to support the finding of liability. Since we find no error, we affirm the judgment of the district court.
The Alouette II helicopter, U. S. registration #N519, was built by appellant SUD in France and sold to Republic Aviation Corporation in 1957.1 It was delivered to Republic in France in February 1959 and shipped to the United States in a partially disassembled state by Republic, who reassembled it and used it for approximately 437 flight hours as a demonstration model. On August 1, 1959, Republic leased it to Petroleum Helicopters, Inc. (PHI) for use in its business of supplying air transportation to companies conducting oil exploration in the Gulf of Mexico. The crash occurred on November 30, 1959, while PHI pilot George Stubbs was ferrying Luke H. Wade, Jr. and Harold J. Nicol, employees of the Gulf Oil Corp., to an off-shore rig in the Gulf of Mexico. All three were killed when the helicopter fell into the water while Stubbs was attempting to return the disabled craft to land.
It is agreed that the crash resulted from a break or “failure" in the structural steel of the helicopter’s tail, on the *430upper right longeron (one of the three steel tubes making up the tail boom) at a point where the horizontal stabilizer bracket was welded to the longeron. At issue is whether the break was caused by a defective weld at the point of failure, or by some other cause for which the manufacturer would not be responsible.
At the trial libelants attempted to prove that the longeron broke because of “incomplete root penetration” in the weld, which caused unusual stress on the longeron and led to the failure. Appellant, on the other hand, maintained that the weld was not defective and that some unusual shock or strain, resulting from mishandling by PHI, caused a crack in the longeron, near the weld, which progressed over a period of time to a complete fracture.
Evidence to support the theory that a defective weld caused the accident came largely from libelants’ expert witnesses, William L. Holshouser, whose deposition was read into the record at length, and Isaac Stewart, who testified in person. Holshouser, a Bureau of Standards metallurgist, deposed that his .examination of the broken pieces revealed insufficient root penetration in the weld around which the break occurred. The root of a weld is the point at which the surfaces of the two pieces of metal being welded come together, at right angles in the present case. Incomplete root penetration occurs when the weld metal does not sufficiently penetrate this area where the surfaces of the metals are in contact. Holshouser testified that incomplete root penetration would result in increased stress concentration at the point where the weld metal is joined to the tubing, near the root of the weld, and would reduce the load which could be transmitted between the horizontal stabilizer bracket and the longeron. Expert Stewart testified, largely from close-up photographs of the broken pieces, that there was definitely lack of root penetration. He testified in detail to the effect that incomplete root penetration in the weld would cause normal stresses originating in the horizontal stabilizer to be sidetracked through the weld metal to a necessarily weaker and more brittle part of the weld area rather than being passed directly through the root of the weld.
SUD’s case consisted mainly of testimony from three experts in support of its theory that it was not a defect in the manufacturing which caused the accident. Two of these experts, Everett Chapman and William Cobey, testified that their examinations of the broken pieces revealed nothing unusual about the weld and that it appeared to be a good weld. They also testified that the break which led to the crash could have been initiated by an incident which occurred about three months before the crash, when the tail rotor blades of the helicopter were accidentally dipped into the water during flight. Although there was testimony from libelants’ witnesses that this tail-dipping was not severe, and that the damage caused thereby was minor, there was evidence that extensive repairs were made after the incident. SUD also brought out that the helicopter operating company, PHI, had not conducted a 600 hour inspection, which would have included an examination of the welding to check for cracks, at the time of the crash, which occurred at 622 hours.2 SUD also introduced testimony concerning the high quality of the welding techniques used in its manufacturing of Alouette helicopters.
Judge Croake, in a detailed opinion, found that libelants “displaced stress” theory presented the most probable explanation of how the break in the lon-geron occurred. In addition to crediting *431Mr. Stewart’s testimony on this matter, the court stated:
“The fact that one breakline ran along the edge of the bracket lends support to the theory that a bending stress was applied that the longeron was unable to withstand. It appears that the bracket may have acted as a fulcrum around which the longeron was bent when the displaced stress was applied at a point on the longeron some distance away from the root of the weld.”
* * * * * *
“The fact that the second breakline runs along the edge of the weld fillet also supports the theory that some of the stress was displaced through the weld metal to the weaker, heat-affected zone.”
The court expressly rejected SUD’s theory that “rough handling” by PHI caused the accident, finding that this explanation had many weaknesses and was not persuasive.3 On the basis of these findings, the court concluded that SUD was liable in both negligence and implied warranty.
On appeal SUD contends that libel-ants did not establish by a preponderance of the evidence that the weld was defective and that the defect in the weld caused the crash, the two elements admittedly essential for a finding of liability based upon negligence or breach of warranty. In support of this position, SUD argues that the district court, rather than basing its finding of liability upon findings that the weld was defective and that this caused the crash, based it upon SUD’s inability to prove that the weld was not defective or that something other than a defective weld caused the accident. In other words, SUD claims that the court erroneously placed the ultimate .burden of persuasion upon it rather than upon libelants.
While we agree with the appellant that the ultimate burden of proof remains on the libelants throughout a case such as this, Swain v. Boeing Airplane Co., 337 F.2d 940 (2 Cir. 1964), we do not agree that the trial court erroneously placed this burden on SUD.
It is true that the court’s discussion of the burden of proof is ambiguous, especially where it speaks of appellant’s “burden of establishing as a cause of the accident some factor unrelated to a defective part.” We think, however, that the opinion read as a whole makes it sufficiently clear that the court accepted libelants’ insufficient root penetration explanation, together with the fact that the failure occurred during normal operating conditions, as establishing that the weld was defective and that this probably caused the crash. This placed upon SUD a burden of going forward with evidence of another possible cause; the rejection of SUD’s evidence left the failure of the defective part as the cause. In short, we think the court left the ultimate burden of persuasion on the libelants, holding that their proof and the inferences drawn from it (together with the absence of any other plausible explanation) established that it was more likely than not that faulty manufacturing caused the accident. Cf. Swain v. Boeing Airplane Co., 337 F.2d 940 (2 Cir. 1964); Prosser, Torts, at 215-216 (3 ed. 1964) 4
*432The evidence adequately supports the court’s findings. There was testimony, which the court credited, that there was a defective weld and that this defect would have caused abnormal stress in the area of the break. This evidence, together with the inferences which the district court drew from it and from the other evidence in the case, provides ample support for holding the manufacturer liable. Cf. United States v. Springfield, 276 F.2d 798 (5 Cir. 1960).
We also find that there was evidence to support the district court’s rejection of SUD’s claim that the accident was caused by rough handling rather than by faulty manufacturing. One of the reasons for rejecting this claim, which we find persuasive, is that the fracture apparently was not a “fatigue” fracture but was a sudden or “brittle” fracture.5 For rough handling to have caused the crash, a crack would probably have resulted from the tail-dipping incident, three months before the crash, and progressed by fatigue until the final failure. In addition, it is apparent from the opinion below that the court credited testimony that the tail-dipping incident was not severe and that the repairs made thereafter were largely precautionary in nature.
We cannot overturn the district court’s factual determinations, which were made after a full trial and which obviously involved the credibility of the witnesses who testified,6 unless on all the evidence we are left with a definite and firm conviction that a mistake has been committed. United States v. United States Gypsum Co., 333 U.S. 364, 396, 68 S.Ct. 525, 92 L.Ed. 746 (1948); Montgomery v. Goodyear Aircraft Corp., 392 F.2d 777 (2 Cir. 1968); Fed.R.Civ. P. 52(a). After reviewing the entire record on appeal, we cannot say that Judge Croake’s findings are clearly erroneous.
Judgment affirmed.