In this paper we propose an unsupervised context-sensitive technique for change-detection in multitemporal remote sensing images. Here a modified Self-Organizing Feature Map Neural Network is used. Each spatial position of the input image corresponds to a neuron in the output layer and the number of neurons in the input layer is equal to the dimension of the input patterns. The network is updated depending on some threshold value and when the network converges status of output neurons depict the change-detection map. To select a suitable threshold for initialization of the network, a correlation based and an energy based criteria are suggested. Experimental results, carried out on two multispectral remote sensing images, confirm the effectiveness of the proposed approach.