In this article, we propose an automatic diabetic retinopathy screening method. In particular, we focus on detecting microaneurysms in retina photographs, as they are the most common and first appearing lesions in the disease development. This is done by matching a lesion template in the wavelet domain, using the sum of the squared errors as a criterion on some decomposition subbands. The method outperforms classification methods in the wavelet domain, which seem unfitted to describe small structures shapes. More, it could be generalized to other small lesions. Results are evaluated on a manually segmented retinal images database for different usual mother wavelets.