The analysis of records of species occurrences on islands in an attempt to detect interactions between species has been an area of controversy in recent years in terms of the validity of some of the statistical methods used. In this note I make two contributions to the continuing debate. First, I advocate a generalized Monte Carlo testing procedure because this is easy to implement, is computationally efficient, and has guaranteed properties when the null hypothesis of no species interactions is correct. Second, I propose a test statistic that can be decomposed into a component for each individual species, and I demonstrate how this makes it possible to isolate species with unusual patterns of co—occurrence with other species, even after an allowance for multiple testing is made.