Colposcope is a device which generates a colposcopic images of the cervix which is vital for detecting cervical neoplasia. Thus, analysis of these images plays a significant role in identifying and diagnosing cervical abnormalities. In this paper, we present a UNET model for automatic segmentation of normal and abnormal regions from colposcopic images. This will be beneficial for swift analysis and minimizes the subjective biases between clinicians. The segmentation works by pre-processing the images, training the UNET model on the set of annotated dataset and finally applying the trained model to predict the normal and abnormal regions. The whole algorithm is tested on Kaggle and ICAR dataset.