The search of an image in image database using keywords is made powerful due to automatic image annotation. In this paper, an automatic image annotation using K- Nearest Neighbor (K-NN) is presented. The categorization based approach is presented for annotation. Images are first segmented using k-means clustering and then processed to form feature vector. Local features are extracted from the regions of the image. The feature vectors are experimented using K-NN. Our system is validated using ten categories from the COREL images. It is observed that in multiple instance learning using K-NN with color and texture features outperforms for all type of feature vectors.