The reliability evaluation of sparse representation-based classifiers is greatly affected by the number of training samples, resulting in a strong evidence conflict problem in case of small samples. A conflict evidence fusion method for sparse representation-based classifiers is proposed based on the tendency of being classified into different categories. The method uses the sparse reconstruction errors to quantify the classification tendency of samples, and modifies the confusion matrix by solving the probability of difference of reconstruction errors to obtain the probability that the samples are classified into each category. The modified confusion matrix is used to weight and fuse the evidence sources to solve the problem of high evidence conflicts caused by the low accuracies of identification in case of small samples. In the experiment of bearing fault fusion diagnosis, this method is applied to fuse the sparse representation-based classifiers established from different eigenvectors of different observation signals. The experimental results effectively verify the advantages of the proposed method in solving the problem of high-level conflict evidence fusion in case of small samples.