Illusions of grammaticality have often been used to probe the properties of the human sentence processor in syntactic activities like subject--verb agreement, reflexive binding, and negative polarity item (NPI) licensing. Originally, NPI licensing in processing was thought to be a product of cue-based retrieval. Mounting evidence that the NPI illusion is far from universal suggests that a revised account is necessary. We examine the distribution of the NPI illusion using a single methodology and evaluate its compatibility with existing theories. We find that most licensors fail to show illusion behavior but the negative quantifier "no" and the quantificational phrase "not a single" trigger illusion in high and low relative clause positions. This evidence indicates that distribution of NPI illusion is not predicted by existing processing accounts. Future explanations must engage the unique properties of negative quantifiers to account for the distribution of the NPI illusion phenomenon.