This paper investigates the effectiveness of sparse representation (SRC) and collaborative representation (CR) in face recognition (FR). It argues that while SRC has been widely used, its success is not due to $ l_1 $-norm sparsity but rather the collaborative representation (CR) mechanism. The study proposes a new classification scheme, CRC_RLS, which uses regularized least squares with collaborative representation, achieving competitive performance with significantly lower complexity than SRC. SRC works by representing a test sample as a sparse linear combination of training samples and classifying based on the minimum representation error. However, the paper shows that the key factor in SRC's effectiveness is CR, not $ l_1 $-norm sparsity. CRC_RLS, which uses $ l_2 $-norm regularization, achieves similar or better performance than SRC with much lower computational complexity. Experiments on several face databases (Extended Yale B, AR, Multi-PIE) show that CRC_RLS outperforms SRC in terms of speed and accuracy. It achieves high recognition rates, even under challenging conditions such as occlusion with sunglasses and scarves. The results indicate that CR is more effective than $ l_1 $-norm sparsity for face recognition. CRC_RLS is also more efficient than other methods like SVM, LRC, and NN. The paper concludes that CR is the key to improving FR accuracy, and CRC_RLS is a more efficient and effective method for face recognition.This paper investigates the effectiveness of sparse representation (SRC) and collaborative representation (CR) in face recognition (FR). It argues that while SRC has been widely used, its success is not due to $ l_1 $-norm sparsity but rather the collaborative representation (CR) mechanism. The study proposes a new classification scheme, CRC_RLS, which uses regularized least squares with collaborative representation, achieving competitive performance with significantly lower complexity than SRC. SRC works by representing a test sample as a sparse linear combination of training samples and classifying based on the minimum representation error. However, the paper shows that the key factor in SRC's effectiveness is CR, not $ l_1 $-norm sparsity. CRC_RLS, which uses $ l_2 $-norm regularization, achieves similar or better performance than SRC with much lower computational complexity. Experiments on several face databases (Extended Yale B, AR, Multi-PIE) show that CRC_RLS outperforms SRC in terms of speed and accuracy. It achieves high recognition rates, even under challenging conditions such as occlusion with sunglasses and scarves. The results indicate that CR is more effective than $ l_1 $-norm sparsity for face recognition. CRC_RLS is also more efficient than other methods like SVM, LRC, and NN. The paper concludes that CR is the key to improving FR accuracy, and CRC_RLS is a more efficient and effective method for face recognition.