A detailed study of ‘supercooling' in tin rods has confirmed the hypothesis that this is caused by the difficulty of forming a nucleus of the superconducting phase. The experiments show that when nucleation does occur it only happens at certain flaws in the metal, which have been proved to lie at the surface and to be between 10
-4
and 10
-3
cm in size. These flaws do not appear to be necessarily associated with surface conditions, impurity content, or crystal boundaries. Any handling of the specimen affects them, but simply warming it to room temperature often does not, in which case the supercooling observed in separate experiments is reproducible. The degree of supercooling
ϕ
l
defined as (
H
2
c
-
H
2
l
)/
H
2
c
, where
H
l
is the field at which nucleation is first possible, varies in magnitude from flaw to flaw (0⋅8 is the largest value found), but it always depends on temperature in the same way, rising as
T
→
T
c
, The supercooling can sometimes be increased by applying a high field to the specimen beforehand, indicating that some of the flaws can be temporarily destroyed by such treatment. The behaviour of the flaws can be accounted for if they are assumed to be domains where the interphase surface tension has become negative, perhaps as the result of local lattice distortion produced by dislocations. Using a simple model based on this picture the magnitude and temperature dependence of
ϕ
l
have been explained quantitatively in terms of the characteristic flaw size and the positive surface tension in the undistorted metal (as estimated from work on the intermediate state). Slight superheating (up to 1⋅5% of
H
c
) has also been observed.