Making classification decisions on the basis of multiple calls from the same class (e.g., species) can reduce the classification error that is attributable to variance. In many cases, clustering acoustic signals from multiple classes on the basis of location information can allow for separation and simultaneous classification of those calls. Source location and call timing are the most common criteria for clustering calls into groups or tracks. However, in many studies, widely spaced arrays or internal clock drift render a large proportion of the detected calls non-localizable. We built flexible clustering systems that incorporate all time difference of arrival values from calls detected by two or more sensors. We developed three clustering algorithms and used Monte Carlo simulations to compare their clustering accuracy and classification error. The algorithms increased clustering accuracy and reduced classification error by 2%–20% relative to grouping by time (a common fusion strategy) or classification of individual calls. These results demonstrate the ability of imperfect spatial information to improve classification accuracy of vocally active species.