Reading tests are assumed to be accurate estimates of premorbid intelligence, based on the belief that reading remains relatively stable following cerebral injury/disease. However, this assumption has been primarily inferred only from studies comparing differences in reading/intelligence measures between neurologically impaired and normal groups. The current study, using within-subject comparisons, compared the longitudinal stability of reading (WRAT-R/3) and intelligence (WAIS-R FIQ) scores for 39 individuals with cognitive dysfunction referred for repeat neuropsychological evaluation. Wilcoxon signed-ranks tests indicated that reading scores: (1) did not statistically change for those demonstrating intellectual decline (>0 point decline in FIQ), or for those who remained relatively stable (FIQ increase between 0 and 6 points) over retest, but (2) did significantly improve for those demonstrating intellectual improvement greater than 6 points. These results suggest that reading scores may be appropriately considered as “hold” tests for individuals who intellectually decline/remain stable over time, but not for those demonstrating significant intellectual improvement. Additionally, significant variability in reading score decline/improvement suggests that caution must be used when estimating premorbid intelligence based on WRAT-R/3 Reading scores. (JINS, 1996, 2, 282–285.)