The standard experimental paradigm for providing constraints on models of visual word recognition is to factorially manipulate a set of variables, while attempting to control for extranious variables that may be related to the dependent measure. There are a number of potential concerns about this approach that have been identified in the recent literature. Item- and subject-level regression analyses will be presented on speeded word naming and lexical decision performance on over 2400 single syllable words for both younger adults and older adults. The results highlight task constraints on the influence of standard predictor variables, such as feedforward and feedback consistency, along with constraints for the viability of available models of lexical processing. Preliminary results will also be presented from a multi-site collaborative study designed to provide naming and lexical decision latencies for over 40,000 words and nonwords.