In this paper, the heteroscedastic semiparametric errors-in-variables (EV) model, yi = xi(i)beta + g(t(i)) + epsilon(i), x(i) = xi(i) + mu(i), 1 <= i <= n, is considered, where epsilon(i) = sigma(i)e(i), sigma(2)(i) = f (u(i)), beta is an unknown parameter to be estimated and g(center dot) and f (center dot) are unknown functions to be estimated. Under some suitable conditions, asymptotic properties for the estimators of beta, g(center dot) and f (center dot) are presented based on alpha-mixing random errors. In addition, finite sample behavior of the estimators is provided via simulations to verify the validity of the theoretical results.