The “five-parameter model” is a performance model for photovoltaic solar cells that predicts the voltage and current output by representing the cells as an equivalent electrical circuit with radiation and temperature-dependent components. An important feature of the five-parameter model is that its parameters can be determined using data commonly provided by module manufacturers on their published datasheets. This paper documents the predictive capability of the five-parameter model and proposes modifications to improve its performance using approximately 30 days of field-measured meteorological and module data from a wide range of cell technologies, including monocrystalline, polycrystalline, amorphous silicon, and copper indium diselenide (CIS). The standard five-parameter model is capable of predicting the performance of monocrystalline and polycrystalline silicon modules within approximately 6% RMS but is slightly less accurate for a thin-film CIS and an amorphous silicon array. Errors for the amorphous technology are reduced to approximately 5% RMS by using input data obtained after the module underwent an initial degradation in output due to aging. The robustness and possible improvements to the five-parameter model were also evaluated. A sensitivity analysis of the five-parameter model shows that all model inputs that are difficult to determine and not provided by manufacturer datasheets such as the glazing material properties, the semiconductor band gap energy, and the ground reflectance may be represented by approximate values independent of the PV technology. Modifications to the five-parameter model tested during this research did not appreciably improve the overall model performance. Additional dependence introduced by a seven-parameter model had a less than 1% RMS effect on maximum power predictions for the amorphous technology and increased the modeling errors for this array 4% RMS at open-circuit conditions. Adding a current sink to the equivalent circuit to better model recombination currents had little effect on the model behavior.