If voters care about the size of the government's majority, then by-election votes should exaggerate national swings. Moreover, if there is uncertainty about the outcome of the general election and if voters” preferences are skewed in such a way as to give more weight to the “downside” outcome (least favourite party wins) than the “upside” (favoured party wins with a larger than preferred majority), then there will be a systematic tendency for governments to lose by-elections, regardless of any changes in national support. These predictions go beyond those generated by conventional explanations. The theory is successfully tested against data from 383 post-War elections.