The Sugeno-type fuzzy logic inference method and the PVSAT-2 method were used to develop models of a photovoltaic system. The models use solar irradiance in the array plane and back-ofmodule temperature in order to predict alternating current (AC) power production. Measured data from a photovoltaic (PV) system in Quebec, Canada was used to train and validate the models. Global models, over all irradiance values, and models developed for different operating regimes dictated by irradiance values were computed. Model validation shows that the Sugeno-type fuzzy logic model outperforms the PVSAT-2 model when sufficient amounts of data are available, with validation root mean square errors of 5.6% and 6.5%, respectively, for the global models. Meanwhile, training separate models over different irradiance regimes yielded no improvement in accuracy with respect to global models. The predictive accuracy of the Sugeno model as a function of the number of inference rules and the effect of the training dataset size on the performance of both methods were studied. The results indicate that the models have relatively good prediction accuracy when trained on a small dataset (1 month), and that the PVSAT-2 is more robust to accommodating observations that are outside the training dataspace. The models developed will form the basis of a fault detection and diagnosis system for PV systems.