We propose a method that improves the constant false alarm rate (CFAR) detection of targets in a nonhomogenous clutter environment. We assume that the clutter data from neighboring cells may have an edge with unknown location dividing data samples into two parts with different independent and identically distributed (IID) Weibull distributions. We propose automatic clutter edge localization which allows us to eliminate the outlier/misleading data, and thereby improves the CFAR detection performance. We also suggest a CFAR detector (CFARD) which uses this estimator.