The invention relates to an improved minimum variance ultrasonic imaging method, belonging to the ultrasonic imaging technical field. The method comprises: performing delay processing and anterior-posterior dimension smoothing processing on sampling signals of receiving array elements to obtain sample covariance matrix estimation; performing eigenvalue decomposition on the anterior-posterior dimension covariance matrix estimation, constructing signal subspaces, and meanwhile adding a pair of constraint conditions to a direction vector by utilizing an ellipsoid covering direction vector value domain and a limitation direction vector module value; in an expected signal subspace, calculating adaptive beamforming weights based on a minimum variance criterion; and employing the adaptive beamforming weights to perform weighted summation on multi-channel data after anterior-posterior dimension smoothing processing to obtain a channel of adaptive beam signals. The method can solve the problems of image resolution, contrast and noise robustness of a present minimum variance algorithm, and improve ultrasonic imaging quality on the whole.