Marked spatial autocorrelation was encountered in an extensive data set on Antarctic seal densities as well as Antarctic pack ice characteristics. Whilst the methodology of measuring spatial autocorrelation is well developed, there is no established infrastructure for statistical inference in terms of correlation analysis or ANOVA. We survey the literature that deals with these problems, as well as some of the approaches that have been proposed for taking autocorrelation into account in inferential statistics. We apply these approaches to a data set comprising Antarctic pack ice seal counts as well as a few environmental measures. In contrast to the predictions from the existing literature, nonlinear estimation suggested that Pearson's r substantially overestimates the true correlation between seal densities and environmental variables. When compared to spatially adjusted analysis of variance, conventional ANOVA that compared seal densities or pack ice characteristics in different areas overestimated the degree of difference between these areas in proportion to the degree of spatial autocorrelation of the particular data set. In our case, the effects of spatial autocorrelation were not neutralised by treating entire transects as single points. These results emphasise the need for a methodology that takes spatial autocorrelation into account for interpreting the spatial data on Antarctic seals collected during the Antarctic pack ice seal (APIS) program. New software written for performing these analyses is available from the senior author.