Abstract : In recent years, automated retina image analysis (ARIA) algorithms have received increasing interest by the medical imaging analysis community. Particular attention has been given to techniques able to automate the pre-screening of Diabetic Retinopathy (DR) using inexpensive retina fundus cameras. With the growing number of diabetics worldwide, these techniques have the potential benefits of broad-based, inexpensive screening. The contribution of this paper is twofold: first, we propose a straightforward pipeline from microaneurysm (an early sign of DR) detection to automatic classification of DR without employing any additional features; then, we quantify the generalisation ability of the MA detection method by employing synthetic examples and, more importantly, we experiment with two public datasets which consist of more than 1,350 images graded as normal or showing signs of DR. With cross-datasets tests, we obtained results better or comparable to other recent methods. Since our experiments are performed only on publicly available datasets, our results are directly comparable with those of other research groups.