Stochastic weather generators have had many applications in climatology, including agricultural risk assessment and generation of regional climate change scenarios. These models typically use Markov chains or wet/dry spell length distributions to simulate precipitation occurrence, gamma or mixed exponential distributions for precipitation amount, and a continuous multivariate stochastic process for daily maximum and minimum air temperature and total daily solar radiation. The generated sequences are constrained to have the desired serial and cross-correlations by two matrices, A and B. However, in many weather-generator implementations, these matrices are treated as constant with respect to location, time of year, and wet/dry status. In this study, the effects of varying these parameterizations are investigated. Using daily weather observations from numerous stations across the contiguous United States, long series of daily weather data are generated using a WGEN-type stochastic weather generator. We examine the spatial and seasonal differences in the values of A and B, as well as the differences between simulated weather series when A and B are held constant and when they are allowed to vary by location, wet/dry status, and time of year. Differences between the resulting weather series are compared graphically and statistically. Impacts of the parameterizations on generated weather sequences are discussed in terms of weather generator applications.