With the widespread use of convolutional neural networks in the field of deep learning, which have demonstrated significant results in several areas, there is an increasing demand for explainable decisions within them. Class Activation Maps(CAM), which highlight the salient regions of the network’s decisions, are a common method for visualizing and analyzing deep learning networks in the field of computer vision. Several methods have been proposed, which generate visual interpretation maps by linear combinations of activation maps from convolutional neural network (CNN). However, most of these methods use gradient visualization techniques, while gradient-free visualization techniques are too inefficient. In this paper, we develop a new a post-hoc visual explanation method called: Miss-CAM, which, unlike previous gradient-based visualization methods, is inspired by receptive field, where the weights of pixels are obtained by the mask to eliminate gradient dependencies. We experimentally demonstrate that Miss-CAM has better visual performance and fairness interpretation of the decision process. Our method outperforms previous methods in generation efficiency, recognition and localization tasks.