Spinocerebellar ataxia type 3 (SCA3), like other polyglutamine (polyQ) diseases, is characterized by the formation of intraneuronal inclusions, but the mechanism underlying their formation is poorly understood. Here, we tested the "toxic fragment hypothesis", which predicts that proteolytic production of polyQ-containing fragments from the full-length disease protein initiates the aggregation process associated with inclusion formation and cellular dysfunction. We demonstrate that the removal of the N-terminus of polyQ-expanded ataxin-3 (AT3) is required for aggregation in vitro and in vivo. Consistently, proteolytic cleavage of full-length, pathogenic AT3 initiates the formation of sodium dodecylsulfate-resistant aggregates in neuroblastoma cells. Although full-length AT3 does not readily aggregate on its own, it is susceptible to co-aggregation with polyQ-expanded AT3 fragments. Interestingly, interaction with soluble polyQ-elongated fragments causes a structural distortion of wild-type AT3 prior to the formation of stable co-aggregates. These results establish the critical role of C-terminal, proteolytic fragments of AT3 in the molecular pathomechanism of SCA3, in strong support of the toxic fragment hypothesis.