A variety of studies have reported correlations between ratings of beauty and ratings of perceived usability ranging from .92 to .00. This may be due to methodological inconsistencies. This paper will propose a methodology capable of providing a more definitive characterisation of the strength of this relationship. The method requires that participants and products be treated similarly. It is standard practice to specify the pool from which participants are drawn (e.g., undergraduate volunteers from such and such a university) and then make a case that the sample is representative of this pool because they are sampled randomly or exhaustively. One should similarly specify the pool from which products are sampled (e.g., health oriented websites) and then sample randomly or exhaustively from that pool. An analysis that aggregates ratings across products and has participants as sampling units is referred to as a subjects analysis. An analysis that aggregates ratings across participants and has product as the sampling unit is referred to as a materials analysis. To claim that a particular pattern of correlations is reliable it must be demonstrated in both subjects and materials analyses. The former demonstrates that the results apply across the defined participant pool and the latter across the product pool.

Two data sets were analysed. For Set 1, 60 participants rated 10 e-commerce websites and ratings were averaged across sites to give a set of scores for 60 participants (subjects analysis). For Set 2, 10 participants rated 60 e-commerce websites and ratings were averaged across participants to give a set of scores for 60 websites (materials analysis). In both cases bivariate correlations between beauty and usability were positive but low. Correlations between beauty and goodness were moderate (subjects analysis) or high (materials analysis). A mediation analysis shows that what correlation there is between beauty and usability is wholly accounted for by the indirect effect of goodness.

An inference perspective on quality perceptions and judgment is developed suggesting that when users are unable to judge usability directly they infer it from goodness which is highly influenced by more rapidly judged attributes such as presentation. This is not found to be the case for another judged product characteristic, that is, hedonic product character where there is a stronger and demonstrably direct correlation.