The most commonly used tools for daily precipitation modeling are the 2-state Markov chain models. We present the results from a monthly analysis of Markov chain models for 831 stations in the contiguous USA using long-term data and discuss the temporal and spatial variations in model order as identified using the Bayesian information criteria (BIC). The maximum likelihood estimates of the Markov transition probabilities are then used to generate 100-year precipitation occurrence series. The distributions of dry and wet spell lengths in the resulting series are then compared to observations using a two-sample Kolmogorov-Smirnov (K-S) test. In general, the model order identified by the BIC usually fails to accurately reproduce the lengths of wet and dry spells according to the K-S test. It is therefore recommended that users of precipitation occurrence models consider models of higher order than those identified by the BIC or use additional or different model order criteria.