Recently, many attempts have been made to measure the difference in velocity between the fastest and slowest fibers in a nerve trunk or to estimate the nerve fiber conduction velocity distribution in a nerve bundle using several different methods, such as collision technique (Hopf 1963) and computer analysis of the compound action potentials (Cummins et al. 1979; Barker et al. 1979). For the computer analysis, however, some assumptions in regard to the quantitative relationship among conduction velocity, single fiber action potential and fiber diameter are necessary, and there has been little agreement about them. There is also a problem about the relationship between conduction velocity and refractory period in Hopf's technique. Using a collision technique with a method of 3-point stimulation, Gilliat et al. (1976) now suggested that surface recording was unsatisfactory for measuring the velocity of the slow-conducting nerve fibers. With this method, however, we had a preliminary experiment to analyze conduction velocities of so-called A fibers in the bullfrog's sciatic-peroneal nerve using fluid electrode, and they were divided into 3 groups (Nakanishi et al. 1986). These findings were in good agreement with those obtained by Erlanger and Gasser (1937) using monophasic recording. Therefore, clinical measurement of the nerve conduction velocities with a method of this collision technique was performed using surface recording.