Our study proposes the use of a free classification task for investigating the dimensions used by listeners in their perception of non-native sounds and for predicting the perceptual discriminability of non-native contrasts. In a free classification task, participants freely group auditory stimuli based on their perceived similarity. These data are analyzed by calculating how often stimuli representing two non-native categories are grouped together by participants. A multidimensional scaling analysis can also be conducted on these similarity data. These results can then be used to predict discriminability and correlated with various acoustic or phonological dimensions to determine the relevant cues for listeners. The viability of this method was examined for both a segmental contrast (German vowels) and a non-segmental contrast (Finnish phonemic length). Perceptual similarity data from the free classification task accurately predicted discriminability in an oddity task and separately provided rich information on how the perceptual space is shaped. These results suggest that a free classification task and related analyses are informative and replicable methods for examining non-native speech perception.