Previous research has shown conflicting findings when subjects have performed some task involving words varying in imagery, part of speech (nouns versus verbs) and word frequency presented to the left and right visual fields. This problem was investigated in the present paper. In the first investigation, a group of subjects rated 308 words for imagery and part of speech. It showed that nouns tended to be of higher imagery than verbs. This implies that it is important to control for part of speech while investigating imagery, and vice versa. Experiments 2 and 3 investigated the effects of imagery and part of speech on recognition performance using different sets of words and subjects while keeping other aspects of the experiments identical. Conflicting results were found when the analysis regarded words as being a 'fixed effect' rather than a 'random effect' (cf. Clark, 1973). A quasi-F analysis revealed no significant effects due to either variable. When the data were combined from the two experiments, a quasi-F analysis revealed that imagery had an equal effect in both visual fields. Thus we can draw the conclusion that the effect of imagery in both visual fields is reliable across both new subjects and new words. However, a sufficiently large sample is required in order to demonstrate this. No other effects were significant. Experiment 4 investigated the effect of frequency on report from the two visual fields. Quasi-F analysis revealed that the effect of frequency was larger in the right visual field than in the left visual field. We can conclude that this effect is reliable across both new subjects and new words--if a sufficiently large sample is used. The findings are discussed. in terms of whether one could ever draw conclusions on the basis of previous work, when the analysis had regarded words as a fixed effect. It was concluded that it was possible to draw tentative conclusions, but a firm conclusion could only be based on an experiment in which large numbers of words had been used and in which the results were significant on a quasi-F analysis.