(Student Award Winning Paper for 1991) Data often contain periodic components plus random variability. Walsh analysis reveals periodicities by fitting rectangular functions to data. It is analogous to Fourier analysis, which represents data as sine and cosine functions. For many behavioral measures, Fourier transforms can produce spurious peaks in power spectra and fail to resolve separable components. Walsh analysis is superior for strongly discontinuous data. The strengths and weaknesses of each transform are discussed, and specific algorithms are given for the newer Walsh technique.There are many methods for analyzing periodicities in data. In the time domain, autocorrelations provide a convenient way to look for cyclic patterns. However, such time-domain analyses often fail to resolve multiple periodicities superimposed in a single data set, particularly if some frequency components have little amplitude relative to others. Frequency-domain analyses typically provide greater resolution and are therefore often a better approach than their time-based counterparts; but many frequency-based methods exist, and the choice of the best method for a given data set is not always straightforward. The purpose of this paper is to compare two methods, the familiar Fourier transform power spectrum and the less familiar Walsh transform power spectrum, and to demonstrate that for some types of data, the Walsh spectrum provides a much clearer and more detailed understanding of the data than does the Fourier spectrum. Algorithms for computing a Walsh spectrum will be presented and explained, and applications of the analysis method will be considered. Figure 1 shows a single trial from a rat pressing a lever for food on a 48-sec fixed-interval schedule in which food was delivered on a random half of the trials. Responses are shown as +Is, and time bins in which no responses occurred are shown as -Is. The trial illustrated is a typical example of responding following omission of food in this procedure; responses occurred in bursts throughout the interval. To determine whether these response bursts were periodically or randomly spaced, it is necessary to choose a method of periodic analysis that will provide an accurate representation of these data.This research was supported in part by grants from the National Institute of Mental Health (R01-MH44234) and the National Science Foundation (BNS 9110158). We would like to thank Russell M. Church for hishelp in exploring these analysis techniques and for aiding in the writing of this manuscipt. We would also like to thank Howard Kaplan for his comments on an earlier draft. Our thanks also to the anonymous reviewers for their comments and suggestions. Correspondence should be addressed