Two groups of 8 young adults (mean IQ = 68), matched on the basis of age and initial estimates of inspection time (lambda), were compared under conditions in which subjects were instructed either to respond as quickly and accurately as possible or to respond accurately and were praised each time that a response was slower than their mean reaction time (RT) during the initial session. Praise slower RT appreciably but did not affect accuracy, so that second measures of lambda were similar to initial estimates. In a second experiment lambda was measured for 8 retarded (mean IQ = 69) and 8 nonretarded adults under conditions in which stimulus-exposure durations either varied randomly or were arranged in regular blocks from long to short durations. Estimates were significantly longer among retarded subjects and were not influenced by procedure. Reaction times were not affected by procedure, being faster on the average for retarded subjects and relatively constant irrespective of exposure duration, whereas for nonretarded subjects RT increased as exposure duration decreased. Despite the resilience of lambda to procedural variation, differences in RT among retarded and nonretarded samples raise questions about which processes lambda is measuring.