Growth models are applicable to mobile telephony diffusion. Although cross-sectional performance comparisons of models are numerous, varying stages of the S-shaped diffusion curve have not been analyzed by longitudinal studies. This study determines whether the best model applies to an entire diffusion life span. Mobile telephone subscriber data for Taiwan during 1988-2007 are analyzed to compare the performance of three popular diffusion models and one well-known forecasting model--the Gompertz, Logistic, Bass, and time-series autoregressive moving average (ARMA) models, respectively. Empirical results indicate that the Gompertz model outperforms the other models before diffusion take-off, and the Logistic model is superior after inflection and over the aggregate range of the diffusion. Network externalities are the dynamics of the Logistic model and account for its excellence. This longitudinal study is the first to present empirical evidence indicating that the appropriate diffusion model for mobile telephony is stage-dependent, complementing the case dependency of the appropriate diffusion model demonstrated by cross-sectional studies.