It is well known that the slope parameters in the linear regression model may be subject to high sampling variance when the regressors are non-orthogonal. A vast number of ridge and shrinkage estimators have been proposed to yield improvements over ordinary least squares or maximum likelihood estimators. The intercept parameter, however, has been given very little attention in the context. We propose a number of intercept estimators for models with non-orthogonal regressors that are based on shrinkage techniques. The optimal values of shrinkage coefficients are obtained according to the minimum mean square error criterion. A good performance of proposed estimators is documented.