Techniques for interactive brushing, as found in multivariate data visualization systems, can be categorized as either screen-space or data-space. Screen-space techniques consider a brush to be a 2-dimensional shape (usually a rectangle) which can be used to select points which map to a particular region of the display. Data-space techniques assume the brush has as many dimensions as the data set, i.e. the brush specifies an N-dimensional subspace of the entire data space. In this paper, I describe methods for specifying and manipulating N-dimensional brushes and show their implementation in the public-domain visualization package XmdvTool. Among the topics covered are user-driven versus data-driven brushing, direct and indirect brush specification and modification, management of multiple simultaneous brushes, composite brush creation, and using brushes with ramped boundaries.