Retinal blood vessel segmentation plays significant role in the diagnosis of fundus diseases like Diabetes retinopathy, Glaucoma, macular degeneration, etc. Numerous algorithm of above said problem has been cited in the literature. Some more attention is required to overcome the shortcomings of these algorithms like database invariability, anti-noise interference ability. This paper proposed an algorithm based on well-known U-Net supported by attention module which will be able to segment the thin blood vessel up to few pixel spans giving more attention to the gradient changes. Proposed method reduced the requirement of memory requirement as number of learnable parameters reduced and it requires less number of training datasets as compared to U-Net. Proposed method has been tested on DRIVE and STARE datasets and achieve the AUC as 0.980 and 0.974 respectively.