Studies of mitotic spindle elongation in vitro using populations of diatom spindles visualized with immunofluorescence microscopy have shown that the two interdigitating half-spindles are driven apart by an ATP-dependent process that generates force in the zone of overlap between half-spindles. To characterize further the system responsible for spindle elongation, we observed spindle elongation directly with polarized light or phase-contrast video-microscopy. We report that the kinetics of spindle elongation versus time are linear. A constant rate of spindle elongation occurs despite the continuous decrease in length of the zone of overlap between half-spindles. The average rate of spindle elongation varies as a function of treatment, and rates measured match spindle elongation rates measured in vivo. When spindles elongated in the presence of polymerizing tubulin (from bovine brain), the extent of elongation was greater than the original zone of half-spindle overlap, but the rate of elongation was constant. No component of force due to tubulin polymerization was found. The total elongation observed in the presence of added tubulin could exceed a doubling of original spindle length, matching the elongation in the intact diatom. The linear rate of spindle elongation in vitro suggests that the force transducer for anaphase B is a mechanochemical ATPase, analogous to dynein or myosin, and that the force for spindle elongation does not arise from stored energy, e.g. in an elastic matrix in the midzone. Additionally, on the basis of observations described here, we conclude that the force-transduction system for spindle elongation must be able to remain in the zone of microtubule overlap during the sliding apart of half-spindles, and that the transducer can generate force between microtubules that are not strictly antiparallel.