Sparsity-based reconstruction methods enable high-precision localization of overlapping images of single molecules. Appropriate dictionary selection is critical for 3D sparsity methods. Important dictionary selection considerations include element step size, generation method, and normalization. OCIS codes: (180.6900) Three-dimensional microscopy, (100.6640) Superresolution, (100.6890) Three-dimensional image processing.In localization microscopy, individual fluorescent probes are localized with high precision. Through a combination of biochemical and optical techniques, one can enforce the condition that only a small fraction of emitters is in an "active" state at any time. In this way, many localizations can be combined from thousands of frames to generate a single image with resolution beyond the limits imposed by diffraction [1-3]. The need for thousands of frames results in low temporal resolution; we address this problem by utilizing localization methods that can tolerate overlapping images of single molecules [4][5][6][7][8][9][10]. More specifically, one can reconstruct a dense scene by exploiting the mathematical sparsity of the scene [5,9,10]. Imposing the sparsity condition requires a dictionary, i.e. a set of functions, that defines a space in which the scene is sparse. Proper dictionary definition and reconstruction enable substantial enhancements in recoverable labeling density and hence acceleration of the data acquisition by one order of magnitude [9]. Here we present several important considerations for generating a dictionary for performing sparsity-based localization.