Music similarity is a multidimensional concept to which so-called "sub-similarities", such as timbre and rhythm similarity, contribute. In this study, two models are presented: one for timbre similarity, and one for rhythm similarity. The musical domain for which the models were established is Electronic Dance Music (EDM). The models extract feature values from segments of audio and calculate a distance between two segments based on their feature vectors. The models are evaluated on perceptual data using linear regression. The accuracy of the rhythm similarity model reaches an empirically established upper bound to model performance. The accuracy of the timbre model is moderate, possibly due to insufficient data. From the selection of features and their weights resulting from the regression analysis, periodicity of rhythmic elements turned out to be the most important feature group for rhythm similarity in EDM.