The modes and probabilities of error in scanning UPC symbols are calculated as functions of noise and printing error. Probabilities of error in character decoding are compared for the different possible modes. A one module error in decoding the T2 interval is seen to be the most likely failure mode. "Convolution distortion" is described for situations where the scanning beam diameter exceeds the width of bars or spaces within the symbol. Examples are given where the probability of a character error increases rapidly with beam diameter. Expressions for detectable and undetectable error probabilities per symbol scan are derived. For Version A symbols, the detectable error rate can be reduced by employing error correction, but at the expense of a higher undetectable error rate. Undetectable error probabilities for Version A symbols are seen to be at least an order of magnitude lower than for Version B and E symbols. The dominant mode for Version A is. the transformation of one-half of the symbol into a Version E symbol. Undetectable error probabilities are shown to increase very rapidly with noise and printing error for all symbol versions.