Detection of defective pixels that develop on-line is a vital part of fault tolerant schemes for repairing imagers during operation. This paper presents a new algorithm for the identification of stuck low, stuck high and partially stuck pixels in both regular and fault tolerant APS systems. The algorithm does not require specialized illuminations but instead operates on a sequence of regular images and uses statistical information extracted from each image to decide the state of each pixel. Unlike previous techniques, simulations of this technique show that it can find all faulty pixels without misidentifying good pixels as faulty. Under typical conditions, the algorithm successfully converges on the correct result within 238 images for a fault tolerant APS, and 16 images for a regular APS. More extensive simulations have shown that these results can be extended to high-resolution sensors and complex defect models that include hot pixels, without a significant decline in performance.