This paper presents a feature detection and concealment algorithm which can conceal errors in DCT coded images. Feature detection, classification and concealment of the lost information is performed totally within the transform domain for such images. The reconstructed image can then be replayed, and the concealment provides a visible improvement to the regenerated image. Only three low frequency DCT coefficients are used in performing the detection and classification process, requiring only a limited amount of computation. The algorithm operates by searching for a single edge in each of the blocks surrounding the area of loss. The edge is then classified according to its angle and displacement within the block. The algorithm then calculates and implements the amount of pixel domain displacement required to produce a block that contains an edge which continues the feature through the error block. This effectively removes the visible discontinuity within the edge or feature produced by the error. This discontinuity is still present if a simpler algorithm is used based on linear interpolation techniques alone. Some results from the application of the algorithm to JPEG images are shown.