This paper proposes a methodology for a statistical analysis of uniaxial compressive strength and applies it to full-scale data collected in the Svalbard area from 2005 to 2011. A total of 894 samples were compressed over 7 years of field investigation. The ice was mainly from frozen fjords on Svalbard and also from the Barents Sea and the Arctic Ocean. The analysis consisted in determining the most appropriate distribution for level ice strength according to the sample orientation, the time of the year (which would then relate to the brine configuration in the ice), and the failure mode. Six groups (horizontal, vertical, early, late, brittle, and ductile) were defined, and the gamma, two-parameter Weibull, and lognormal distributions were compared for each group. The Weibull parameters (shape and scale) were estimated with the method of moments and the method of maximum likelihood. The two methods agreed well. A visual observation of quantiles–quantiles plots (QQ-plots) combined with a linear regression and a Kolmogorov–Smirnov (KS) test were conducted to determine the best fitting distribution. Neither the season nor the failure mode appeared to influence the determination of a statistical distribution, contrary to the sample orientation. However, it appeared that the lognormal distribution was a best fit for the failure mode and season groups, whereas the gamma and the Weibull were the best candidates for the vertical and horizontal samples, respectively.