Multivariate data in ecological applications most often occur in the form of counts of species abundances in assemblages, where each species is a variable. These data do not generally conform to traditional statistical assumptions, and so special approaches and methods are needed in this context. Statisticians need to be informed about these special problems with ecological data. In addition, the rationale for complex experimental designs that is a trademark of most ecological studies needs to be well understood by applied statisticians in this area. On the other hand, successful approaches for teaching ecologists about the use of multivariate statistics include sticking to the conceptual, rather than the mathematical. I provide here an overview of the methods that have helped teaching across these two disciplines, including a general approach for the use of novel non-parametric methods in the analysis of ecological community data.