During the past decade, compositional analysis (CA) has been used widely in wildlife habitat and resource selection studies. However, critical aspects of CA have not been tested for potential systematic biases such as an inflated Type I error rate. We used computer-simulated data based on known habitat use and availability parameters and found that compositional analysis could result in large Type I error rates. These inflated Type I errors occurred when available habitat types that were not used by animals were included in the resource selection analysis. These error rates arise because of the recommended substitution of an arbitrarily small value, such as 0.01, for each 0% utilization value for any animal. We observed, based on a series of computer-simulation analyses, that progressively larger Type I error rates in CA resulted from substituting progressively smaller positive values for each 0% utilization of a habitat category. The Type I error rate in CA also increased when the number of experimental animals was increased for a fixed number of observations per animal. Two other resource selection analysis methods (Neu et al. [1974] and the Euclidean distance-based analysis [DA] method of Conner and Plowman [2001]) did not exhibit inflated Type I error rates for the same simulated data. Our computer simulations cause us to question the veracity of CA habitat selection analyses that include habitat patches or categories with relatively small areas of availabilities and 0% use.