High levels of connected PV generation in distribution systems impact the power quality and network operation considerably. Therefore, it is important to determine the maximum capacity of PV generation that can be installed using appropriate techniques to model the variability of solar resources. This work presents a probabilistic methodology to determine the PV hosting capacity in a radial distribution system using Gaussian Mixture Model and Monte Carlo simulation to characterize the availability of irradiance and the power demand behavior. The maximum allowed PV capacity is determined by analyzing the results of the power flow simulations in terms of the currents, voltages and power factor. The methodology was applied to a real medium voltage distribution grid located in Bogota-Colombia using a Matlab-OpenDSS interface, and its hosting capacity was determined in terms of the maximum voltage variations allowed by the Colombian regulations.