This paper tests between fads and bubbles using a switching regression to distinguish between competing models. Two main features of the bubbles model distinguish it from the fads model. First, the bubbles model implies that returns are drawn from regimes which differ in the way returns vary with deviations from fundamental prices. Second, the bubbles model implies that deviations from fundamental price will help predict regime switches. Using US data for 1926–89, we find evidence which is consistent with the fads model even when we allow for variation in expected dividend growth rates and expected discount rates. However, the restrictions which the fads model implies for a more general switching-regression specification are rejected. The rejections point in the direction of the bubbles model, although not all of the implications of the bubbles model are supported by the data.