This paper examines empirically the relationship among the stability of security and portfolio betas and (1) the length of the sample period used to calculate betas, (2) beta adjustment techniques, and (3) beta magnitudes. Beta values are forecast using four models: (1) a naive model which assumes the beta value in period t + 1 is the same as in period t, (2) Blume's regression model, (3) a regression model used by Merrill Lynch, Pierce, Fenner and Smith, and (4) a Bayesian procedure suggested by Vasicek.