A technique for signature based diagnosis using windows of different sizes is presented. It allows to obtain increased diagnostic information from a given test at a lower cost, without additional hardware. Existing techniques that use signature based methods are limited by occurrences of aliasing that can lead to failure in the diagnosis process. The new approach proposed in this paper uses windows of different sizes based on the distribution of faults in a circuit and reduces the probability of aliasing in a window. Signature analysis can then give reliable information about failing and non-failing vectors. The effectiveness of the proposed method is evaluated by experiments conducted on ISCAS benchmark circuits. The results show that the proposed method can improve the diagnostic resolution and can reduce the cost of diagnosis.