False memories have a tendency to persist over time, while true memories decline. Research has also shown that: (1) false memories generated using the Deese/Roediger-McDermott (DRM; 1995) paradigm are more effective than true memories at priming insight-based problem solutions (compound remote associates task or CRAT) following a one-week delay; and (2) when valence is manipulated, false-negative memory rates increase over a delay interval. The current study examined the efficacy of true and false memory primes for positive and negative DRM lists on a set of CRATs. Two hundred and seventy participants studied either positive- or negative-themed DRM lists whose critical lures were also the solutions later CRATs that they attempted to solve either immediately or one week later. The critical lures were either generated by the participants (false memory) or included as part of the list (true memory). We found that false-negative memories were more effective problem-solving primes than true or positive memories particularly following a one-week delay.