In this work, we establish exponential inequalities which allow us to construct confidence regions, characterized by ellipses, for the least squares estimates of the parameters in the case of a linear regression model with α-mixing errors.