The invention discloses a non-reference quality evaluation method based on a visual cortex orientation selectivity mechanism, and mainly solves problems of low accuracy and poor stability of evaluation in the prior art. The method includes steps: 1, selecting experiment samples from an image database; 2, designing an image local area structural descriptor; 3, calculating an image level-one mode vector by employing the image local area structural descriptor; 4, performing dimension reduction on the level-one mode vector to obtain a level-two mode vector; 5, clustering the level-two mode vector to obtain a mode dictionary; 6, extracting training sample characteristic vectors by employing the mode dictionary; 7, establishing a prediction model by employing the training sample characteristic vectors; 8, extracting test sample characteristic vectors; 9, calculating quality values of the test samples by employing the test sample characteristic vectors and the prediction model; and 10, determining the quality of the test samples according to the quality values of the test samples. According to the method, the accuracy and the stability of quality evaluation are greatly improved, and the method can be applicable to an image processing system regarding optimization of the visual quality as the target.