This study compared the values and reliability of peak running speeds from incremental treadmill protocols with different speed increments. Thirty-one men, recreational, endurance-trained runners performed, in an alternate order, three continuous tests with different speed increments (0.5, 1 and 2 km center dot h(-1)). The tests were repeated using the same order. Peak speed was identified as (1) the greatest speed that could be maintained for a complete minute (Vpeak-60s), (2) the speed of the last complete stage (Vpeak-C) and (3) the speed of the last complete stage added to the product of the speed increment and the completed fraction of the incomplete stage (Vpeak-P). The Vpeak-P was the most reliable, independent of the incremental test protocol (1.5% <= CV <= 1.8%), and differently from Vpeak-C (1.8% <= CV <= 2.6%) and Vpeak-60s (1.6% <= CV <= 3.8%), the Vpeak-P was only slightly influenced by the incremental test protocol. The results suggest that Vpeak-P should be used to assess aerobic capability and monitor training effects and that other determinations of peak speed should be avoided.Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES)