The conventional linear minimum mean square error (LMMSE) estimator is commonly implemented through the sample covariance matrix. This estimator can only be implemented if the sample size N is higher than the observation dimension M. Moreover, this estimator performs poorly when the sample size is not sufficiently large. To address this problem, we propose a new shrinkage LMMSE estimator. The proposed estimator performs efficiently over a wide range of observation dimensions and sample sizes. In contrast to existing methods, the proposed estimator can be applied if M ≥ N. Even if M <; N, the proposed estimator performs more efficiently than existing estimators.