The motor protein kinesin couples a temporally periodic chemical cycle (the hydrolysis of ATP) to a spatially periodic mechanical cycle (movement along a microtubule). To distinguish between different models of such chemical-to-mechanical coupling, we measured the speed of movement of conventional kinesin along microtubules in in vitro motility assays over a wide range of substrate (ATP) and product (ADP and inorganic phosphate) concentrations. In the presence and absence of products, the dependence of speed on [ATP] was well described by the Michaelis–Menten equation. In the absence of products, the K M (the [ATP] required for half-maximal speed) was 28 ± 1 μM, and the maximum speed was 904 nm/s. P i behaved as a competitive inhibitor with K I = 9 ± 1 mM. ADP behaved approximately as a competitive inhibitor with K I = 35 ± 2 μM. The data were compared to four-state kinetic models in which changes in nucleotide state are coupled to chemical and/or mechanical changes. We found that the deviation from competitive inhibition by ADP was inconsistent with models in which P i is released before ADP. This is surprising because all known ATPases (and GTPases) with high structural similarity to the motor domains of kinesin release P i before ADP (or GDP). Our result is therefore inconsistent with models, such as one-headed and inchworm mechanisms, in which the hydrolysis cycle takes place on one head only. However, it is simply explained by hand-over-hand models in which ADP release from one head precedes P i release from the other.