This paper investigates the influence of cloud model statistics on the accuracy of statistical multiple-frequency liquid water path (LWP) retrievals for a ground-based microwave radiometer. Statistical algorithms were developed from a radiosonde data set in which clouds were modeled by using a relative humidity threshold and a modified adiabatic assumption. Evaluation of the algorithms was then performed by applying the algorithms to four data sets in which clouds were generated in different ways (i.e., threshold method, gradient method, and cloud microphysical model). While classical two-channel algorithms, in this case using frequencies at 22.985 and 28.235 GHz, do not show a significant dependency on the cloud model, the inclusion of an additional 50-GHz channel can introduce significant systematic errors. The addition of a 90-GHz frequency to the two-channel algorithm leads to a larger increase in LWP accuracy than in case of the 50-GHz channel and is less sensitive to the choice of cloud model. A drizzle case from the cloud microphysical model shows no significant loss of accuracy for the microwave radiometer algorithms, in contrast to simple cloud radar retrievals of liquid water. In case of rain, however, the results deteriorate when the total liquid water path is larger than 700 g m−2.