We present a method that scans a random field for localized clusters while controling the fraction of false discoveries. We use a kernel density estimator as the test statistic and correct for the bias in this estimator by a method we introduce in this paper. We also show how to combine information across multiple bandwidths while maintaining false discovery control.