The blockbuster chemotherapy drug paclitaxel is widely presumed to cause cell death in tumors as a consequence of mitotic arrest, as it does at concentrations routinely used in cell culture. However, we determine here that paclitaxel levels in primary breast tumors are well below those required to elicit sustained mitotic arrest. Instead, cells in these lower concentrations of drug proceed through mitosis without substantial delay and divide their chromosomes on multipolar spindles, resulting in chromosome missegregation and cell death. Consistent with these cell culture data, the majority of mitotic cells in primary human breast cancers contain multipolar spindles after paclitaxel treatment. Contrary to the previous hypothesis, we find that mitotic arrest is dispensable for tumor regression in patients. These results demonstrate that mitotic arrest is not responsible for the efficacy of paclitaxel, which occurs due to chromosome missegregation on highly abnormal, multipolar spindles. This mechanistic insight may be used to improve selection of future anti-mitotic drugs and to identify a biomarker with which to select patients likely to benefit from paclitaxel.