This systematic review provides a synthesis of intervention research that taught mathematical word problem solving skills to students with autism spectrum disorder between 1975 and April of 2020 by evaluating the body of research in terms of "what works", "for whom" and "under what conditions". The Council for Exceptional Children (CEC, 2014) quality indicators were used to evaluate methodology of the 20 included studies. The 18 studies that met the criteria to be classified as "high quality" were further analyzed in terms of intervention components (i.e., the "what"), how effectiveness was measured (i.e., defining "works"), characteristics of individuals included in the research (i.e., "for whom") and the tasks, settings, and instructors used (i.e., "under what conditions"). While six practices met the CEC criteria for classification "evidence based", including task analysis, system of least prompts, graphic organizers, explicit instruction, schema-based instruction, and technology assisted instruction, these practices were consistently used in combination as "treatment packages". Implications for practice and future research are discussed.