The performance of a mean-level detector is considered when one or more interfering target returns is present in the set of cells used in estimating the reference level. A serious degradation of detection probability is demonstrated for Swerling target fluctuation models 1 and 3. To alleviate the problem we analyse an alternative procedure. A censoring scheme is proposed whereby samples exceeding an adaptive threshold are excluded from the reference set. An iterative procedure is described where samples exceeding a threshold computed with a predetermined scale constant are excluded from the reference set; then the threshold is recomputed using another predetermined scale constant with the censored sample set. The procedure is repeated and eventually terminated when no reference cell sample exceeds the computed threshold. This then forms the detection threshold. Expressions for false-alarm and detection probabilities are derived for general background distribution. Numerical results are presented for the case of Gaussian white noise. The procedure is shown to maintain acceptable performance in dense target environments, providing a significant improvement in the detection probability.