We compare four procedures for estimating new product diffusion models, viz., ordinary least squares, maximum likelihood estimation, nonlinear least squares, and algebraic estimation. The paper first discusses the advantages and disadvantages of the four estimation procedures. An empirical comparison is then made using seven data sets. The results indicate that the nonlinear least squares procedure provides better one-step ahead forecasts. This conclusion, based on the Bass diffusion model, holds up for the two other diffusion models tested, viz., the Mansfield model and the Gompertz curve.