Abstract. Female preferences for male mating signals are often evaluated on single parameters in isolation or small suites of characters. Most signals, however, are composites of many individual parameters. In this study we quantified multivariate traits in the advertisement call of the tú ngara frog, Physalaemus pustulosus. We represented the calls in multidimensional scaling space and chose nine test calls to represent the range of population variation. We then tested females for phonotactic preference between calls in each pair of the nine test calls. We used statistics developed for paired comparisons in such ''round robin'' competitions to evaluate the null hypothesis of equal attractiveness, and to examine the degree to which females responded to calls as being different from or similar to one another in attractiveness. We then examined the attractiveness of each test call relative to all other test calls as a function of their location in multivariate acoustic space (the acoustic landscape) to visualize sexual selection on calls. Finally, we used methods from cognitive psychology to illustrate the females' perception of call attractiveness in multivariate space, and compared this perceptual landscape to the acoustic landscape of quantitative call variation.We show that correlations between individual call characters are not strong and thus there are few biomechanical constraints on their independent evolution. Most call variables differed among males, and there was high repeatability of call characters within males. Females often discriminated between pairs of calls from the population, and there were significant differences among calls in their attractiveness. Female preferences for calls were not stabilizing. The region of the acoustic landscape that was most attractive to females included the mean call but was not centered around it. The females' perceptual or preference landscape did not correlate with the call's acoustic landscape, and female perception of calls decreased rather than enhanced call differences. Jang and Greenfield 1998). Some of these studies have measured preference functions for single traits or pairs of traits and show how the strength of preference, which is sometimes translated to a selection gradient, covaries with the stimulus (Ritchie 1996;Wagner 1998;Murphy and Gerhardt 2000;Schul and Bush 2002).The approaches used above have been invaluable in elucidating how female mating preferences (and sometimes male mating preferences) can generate selection on individual or small suites of signal traits. There are several limitations to these approaches, however. One limitation is that the female preferences for individual traits might not predict how females respond to comparisons of entire signals. The interaction effect of individual traits on female preference might not be additive, and it could even reverse signs. As a specific example, studies of jungle fowl show that females attend to a suite of characters but the relative importance of each character can vary with time (Zuk et ...