Multiple accounts have been proposed to model the mental lexicon, and these models have mainly been validated by comparing how well they can describe human data derived in an experimental setting. Alternatively, we present two experiments where word stimuli were generated using different models and compare their quality in a familiar word guessing game. A text-corpus-based model and a word association model were used to generate lists of closely related words to a target word. Consecutively, these words were used as hints in a word guessing game where the aim was to find the target word. Both models succeeded in generating good quality hints, but those generated from the word association model led to more accurate responses and fewer hints needed. Potential explanations for why the word association model provided a better approximation of the mental lexicon are discussed.