Jones (1976) and Ozaki and Oda (1978) independently introduced a class of nonlinear models known as amplitude-dependent exponential autoregressive (EXPAR) models. Many authors have discussed the usefulness of these models (e.g. Qzaki, 1993). The conditional least squares method has been used frequently to get point estimates for the unknown parameters of these models. However, problems have been raised because of the exponential regression type of the EXPAR models. In this paper, the bootstrap algorithm (Efron, 1979) is employed to estimate the standard errors of the conditional least squares estimates and construct 100 (1-a) % confidence intervals for the unknown parameters of the exponential autoregressive (EXPAR) models. Simulation results are presented to motivate using the bootstrap procedure. A real example, using the famous Canadian lynx data, is given.