Using an international Thompson Datastream database and standard asset pricing models we encounter pricing errors for the ten percent smallest stocks. We generalize the standard 4-factor model by adding two additional risk factors (one size- and one book-tomarket factor). This generalized 6-factor model is tested both on US and international data (with 39 countries both developed and emerging) and is able to price the entire size spectrum. We discuss the possible economic explanations of these risk premia for the smallest stocks. The fact that pricing errors are resolved by adding factors rather than characteristics, rules out data problems and information asymmetries as an explanation. Thin trading bias in the beta is also rejected as the source of abnormal returns. Liquidity remains a serious possible candidate, as is the hypothesis of additional downside risk for the smallest firms.