Well-established rules of translational initiation have been used as a cornerstone in molecular biology to understand gene expression and to frame fundamental questions on what proteins a cell synthesizes, how proteins work and to predict the consequences of mutations. For a group of neurological diseases caused by the abnormal expansion of short segments of DNA (e.g. CAG•CTG repeats), mutations within or outside of predicted coding and non-coding regions are thought to cause disease by protein gain- or loss-of-function or RNA gain-of-function mechanisms. In contrast to these predictions, the recent discovery of repeat-associated non-ATG (RAN) translation showed expansion mutations can express homopolymeric expansion proteins in all three reading frames without an AUG start codon. This unanticipated, non-canonical type of protein translation is length-and hairpin-dependent, takes place without frameshifting or RNA editing and occurs across a variety of repeat motifs. To date, RAN proteins have been reported in spinocerebellar ataxia type 8 (SCA8), myotonic dystrophy type 1 (DM1), fragile X tremor ataxia syndrome (FXTAS) and C9ORF72 amyotrophic lateral sclerosis/frontotemporal dementia (ALS/FTD). In this article, we review what is currently known about RAN translation and recent progress toward understanding its contribution to disease.