One possible avenue that may improve design against buckling is to rec
ognise and account for the random nature of initial geometric imperfec
tions introduced by manufacturing. This paper presents the application
of a probabilistic methodology to the design and analysis of cylindri
cal shells under axial compression. Results from two cases are present
ed and compared. the first involves stringer-stiffened steel cylinders
failing elastoplastically, whereas the second examines unstiffened co
mposite cylinders buckling elastically. In both cases, the method is u
nderpinned by statistical analysis of imperfections measured on nomina
lly identical specimens. Nonlinear FE analysis is used for strength as
sessment and the results of the statistical analysis are introduced in
the imperfection modelling. It is demonstrated that the method has ad
vantages over code design based on 'lower bound' curves, in terms of t
he calculated buckling loads but also in offering a systematic and rat
ional way by which randomness in imperfections can be assessed.