Identification of bat species based on analysis of echolocation calls can be affected by the way data are manipulated, the diversity of species, and call variability. We document the effects of sample sizes and a priori assignment of calls by species on the outcome of discriminant function analysis (DFA) and multinomial logistic regression (MLR) of features of echolocation calls, and determine which features of calls are most useful for identification. We used recorded echolocation calls of eight species readily distinguishable by call features, including molossids, emballonurids and a moormopid recorded at sites in Belize, Brazil, and Mexico. On individual calls, we measured four features: frequency with most energy, highest and lowest frequencies and call durations obtained from sequences consisting of 10 calls. Cluster analysis and multiple analyses of variance indicated significant differences between the calls of different species. Outcomes of DFA and MLR were affected by both sample sizes (numbers of calls, numbers of sequences) and the subjective approach that researchers take to their data (i.e., categorizing calls or sequences of calls by species). Levels of variation in calls of some species in our sample often precluded the use of single calls in making call-based identifications. Accurate documentation of variability in echolocation behavior of sympatric bats is a prerequisite for an effective sound-based bat survey.