Purpose. This study focuses on two psychological mechanisms that may inadvertently affect judges' decisions on proof of guilt and on punishment. It involves mechanisms that are clearly in conflict with formal judicial doctrine. One hypothesis, the conviction paradox, asserts that, faced with very serious offences, a judge's standard of proof will be lower than for less serious, but otherwise comparable, offences. A second hypothesis, compensatory punishment, asserts that in cases with relatively weak evidence, judges who nevertheless render a guilty verdict will be inclined to compensate their initial doubt on the matter of guilt by meting out a less severe sentence. Method. The hypotheses are evaluated in an experiment with Dutch judges and justices who serve in criminal courts. This was done using fictitious but highly realistic dossiers of criminal cases. Results. Neither of the two hypotheses was supported in the present study. Conclusions. Findings are discussed in relation to their implications for theory development and future research in the area of legal decision making. © 2007 The British Psychological Society.