Factors which control lava flow length are still not fully understood. The assumption that flow length as mainly influenced by viscosity was contested by Walker (1973) who proposed that the length of a lava flow was dependent on the mean effusion rate, and by Malin (1980) who concluded that flow length was dependent on erupted volume. Our reanalysis of Malin's data shows that, if short duration and tube-fed flows are eliminated, Malin's Hawaiian flow data are consistent with Walker's assertion. However, the length of a flow can vary, for a given effusion rate, by a factor of 7, and by up to 10 for a given volume. Factors other than effusion rate and volume are therefore clearly important in controlling the lengths of lava flows. We establish the relative importance of the other factors by performing a multivariate analysis of data for recent Hawaiian lava flows. In addition to generating empirical equations relating flow length to other variables, we have developed a non-isothermal Bingham flow model. This computes the channel and levee width of a flow and hence permits the advance rates of flows and their maximum cooling-limited lengths for different gradients and effusion rates to be calculated. Changing rheological properties are taken into account using the ratio of yield strength to viscosity; available field measurements show that this varies systematically from the vent to the front of a lava flow. The model gives reasonable agreement with data from the 1983–1986 Pu'u Oo eruptions and the 1984 eruption of Mauna Loa. The method has also been applied to andesitic and rhyolitic lava flows. It predicts that, while the more silicic lava flows advance at generally slower rates than basaltic flows, their maximum flow lengths, for a given effusion rate, will be greater than for basaltic lava flows.