A widespread problem in the study of animal vocalizations is evaluating the acoustic similarity of signals both between individuals of a social group and between social groups. This problem becomes especially salient when classifying the narrow-band frequency-modulated signals, such as whistles, found in many avian and mammalian species. Whistles are usually characterized by their relative change in frequency over time, known as whistle ‘contour’. Measuring such a characteristic is difficult as it is not a single measurement, such as the mean frequency or duration of a signal, but several associated measurements of frequency across time. This paper reports on a new quantitative technique for determining whistle types based on whistle contour similarity and an application of this technique to the whistles of bottlenose dolphins to demonstrate its utility. This ‘contour similarity’ technique (CS technique) uses cluster analysis to group the correlation coefficients of frequency measurements from a data set of signals. To demonstrate the efficacy of this CS technique, three data sets were analysed, two using computer-generated signals and a third using adult bottlenose dolphin whistles, to (1) examine the efficacy of correlation coefficients for grouping signals by their similarity in whistle contour and (2) determine the viability of this technique for categorizing bottlenose dolphin whistles. Measured actual frequencies and correlation matrices from the four simulated signal types and a correlation matrix from the whistles of five captive adult bottlenose dolphins were each subjected to K-means cluster analysis and the resulting signal types were evaluated. Results indicated that the technique grouped actual frequencies according to the amount of shared actual frequencies and grouped correlation coefficients successfully according to signal contour. This result endured even if contours differed in overall duration or actual frequency or were expanded or compressed with respect to frequency or time. The results suggest that this approach is a viable method for assigning whistle contours to categories in bottlenose dolphins or any other species with narrow-band, frequency-modulated signals.