While numerous researchers have proposed dierent models to forecast trial sales for new products, there is little systematic understanding about which of these models works best, and under what circumstances these ยฎndings change. In this paper, we provide a comprehensive investigation of eight leading published models and three dierent parameter estimation methods. Across 19 dierent datasets encompassing a variety of consumer packaged goods, we observe several systematic patterns that link dierences in model speciยฎcation and estimation to forecasting accuracy. Major ยฎndings include the following observations: (1) when dealing with consumer packaged goods, simple models that allow for relatively limited ยฏexibility (e.g. no S-shaped curves) in the calibration period provide signiยฎcantly better forecasts than more complex speciยฎcations; (2) models that explicitly accommodate heterogeneity in purchasing rates across consumers tend to oer better forecasts than those that do not; and (3) maximum likelihood estimation appears to oer more accurate and stable forecasts than nonlinear least squares. We elaborate on these and other ยฎndings, and oer suggested directions for future research in this area.