We propose a novel probabilistic model for the interference power distributions in a cellular network employing MIMO beamforming in mmWave spectrum. Considering both line-of-sight (LOS) and non-line-of-sight (NLOS) propagations, we use the Gamma distribution for LOS interference power, and propose a mixture of the Inverse Gaussian and Inverse Weibull distributions for the NLOS interference power. For the NLOS mixture model, an expectation maximization algorithm is developed using both analytical moment matching and maximum likelihood estimation. The composite probabilistic model is tested against simulation data obtained from a stochastic geometry based cellular network and demonstrates low relative entropy (Kullback-Leibler distances) for a wide range of practical mmWave path loss and shadowing values.