In this work, we fit multiple non-Gaussian models to high b-value DW-MRI (b= 200-3000 s/mm2) in benign and malignant breast lesions. Models included bi-exponential, stretched exponential, diffusion kurtosis, Padé exponent and ADC (for comparison). We evaluated the quality of fit for each model and investigated the lesion differentiation accuracy for all extracted model parameters. The bi-exponential model provided statistically significant better quality of fit than all the other models, and without systematic bias. This suggests the existence of two effective diffusion compartments in breast lesions. Several of the extracted model parameters performed equally well in terms of lesion differentiation (AUC>0.97).