Source attributions based on positive versus negative valence were examined in four experiments. The two sources were individuals who were depicted positively or negatively, and the content of their statements was similarly valenced. When valenced information about the sources was provided after learning the statements, test biases to attribute positive statements to the positive source and negative statements to the negative source were strongly present. Providing the same information prior to learning improved memory, but did not entirely eliminate test biases (...) based on valence. Signal detection analysis suggests that these “halo effect” biases are criterion-based and not memory-based. Therefore, the results are more consistent with descriptions of source-monitoring processes that can benefit from familiarity-based partial information as opposed to descriptions in which source monitoring is primarily recollection-based. (shrink)