Risk assessment for mixtures of environmental chemicals would be prohibitive if each closely related mixture (e.g., those with components in differing proportions) required separate evaluation. Absent suitable data on a target mixture, the U.S. Environmental Protection Agency considers surrogate data from a "sufficiently similar" mixture for assessing risk. While "sufficient similarity" is imprecisely defined, it does imply a metric whereby mixtures can be declared 'close enough' for risk assessment. Studies that seek to operationalize the idea have used data on chemical components, sources and sampling sites, as well as data from dose-response studies. Proposals involving dose-response data, largely restricted to a single outcome, fit parametric models and define a relevant metric through estimated parameters. Instead, motivated by studies of over-the-counter botanical preparations conducted by the National Toxicology Program, we focus on a model-free approach that uses dose-response data for multiple outcomes simultaneously. We combine multidimensional scaling and hierarchical clustering to identify groups of sufficiently similar preparations.