Rapid advances in technologies have enabled scientists to measure the molecular composition of individual cells with increasing detail and throughput. As these technologies have become widely adopted, they have exposed widespread molecular variability even among cells previously thought to be identical. However, despite the excitement over these discoveries, for many scientists a fundamental question remains: what forms of variability matter for differences in single-cell behavior? Here we describe the development and application of two methodologies for connecting the molecular profile of a cell (cell state) with its future behavior (cell fate), with particular applications for rare biological phenomena. Our first approach combines the experimental design of Luria and Delbrück’s classic “fluctuation analysis” with modern RNA sequencing techniques to identify groups of genes that are coordinately expressed in rare cells and whose expression persists through multiple cell divisions. Applied to multiple cancer models, we identify and validate several such gene expression programs and furthermore, demonstrate that the rare cell subpopulations marked by these programs are far more likely to survive drug treatments. Our second methodology searches for functional forms of single-cell variability from the opposite direction, starting with the unique behavior and effectively going back in time to isolate the cells from which it originated. Combining transcribed single-cell barcodes with high-sensitivity RNA FISH, this methodology is able to selectively recover cells as rare as 1:10,000 (from a population of millions) which can then be profiled using routine sequencing- or imaging-based assays. Using this approach in the context of therapy resistance in cancer, we uncover a variety of resistance outcomes that can be traced back, through weeks of selection and growth, to previously hidden axes of variability in the initial cell population before treatment. These findings begin to detail the complex mapping between rare-cell behaviors in cancer and the underlying molecular variability that enables these behaviors. Moreover, our work outlines a general strategy to uncover such mappings, with likely implication for all manner of phenomena in cancer and beyond.