We propose a algorithm to classify the retinal images as normal or abnormal based on the candide extraction. The image is first preprocessed to remove the noised in them. For preprocessing we use two concepts Illumination equalization and histogram equalization. Then the candide extraction is done to find the abnormality in the preprocessed images. After candide extraction the features such as area, mean, Standard deviation are extracted. These are all done for the query image and the value is stored as test features. The same procedure is repeated for all the images in the database and the features are stored as Train image features. For all the train images true label is set. The Test image feature Train image feature and the true labels were passed into the classifier. The Classifier matches the test image features with the train image features and found the True label and give us the classification result. We use ANFIS classifier in our paper. Finally the performance of the classifier is analyzed.