JPEG-compressed images, when transmitted over a noisy channel, are highly susceptible to errors. The effects of bit errors on images compressed by any method, including JPEG, are significantly augmented since the majority of statistical redundancies are removed in the compression process. Furthermore, errors affecting the run-lengths and Huffman codewords within JPEG have a much more profound influence on reconstructed images than those affecting only coefficient values. This paper presents a detailed investigation of uncompressed and JPEG-compressed (baseline and progressive) image transmission in noisy channels. Several modifications within JPEG are proposed to improve reconstructed image quality. Among the techniques investigated, a hybrid technique, combining the concepts of the baseline and progressive JPEG modes, appears to give the best performance. Here, the DC coefficients (with absolute and/or fixed bit encoding), AC codewords, and some positional information are separately coded and protected using forward error correction. The actual AC coefficient values are sent separately. This partitioning of codewords and values enable isolating, for special protection, portions of the representation that are highly sensitive to errors and that may cause lengthy error propagation. It enables sending `view-quality' images in the presence of a channel bit error rate as high as 0.01.