This paper investigates the forecast performance of a traditional cross-classification model and alternative models that seek to address the shortcomings of traditional cross-classification analysis, specifically when it has cells with inadequate data. The study uses five cross-sectional datasets collected in the San Francisco Bay Area in 1965, 1981, 1990, 1996, and 2000. Alternative models, estimated with travel data collected in the base year, were assessed for their ability to replicate the number of trips made by households in each cell of a cross-classification matrix and at the traffic zone level, respectively, in each of the five years. The results showed that the traditional crossclassification analysis (CCA) model, notwithstanding having a few unreliable cells provided more consistent predictions of travel than any of the alternative methods. They also show that it is better to synthesize trip rates for only those cells of the cross-classification matrix with inadequate data rather than to adjust the entire trip-rate matrix as is currently the practice.