Five ground-based VHF receivers were used to trace paths followed by flashes of lightning. Every lightning flash radiates a succession of radio noise pulses, and three Cartesian corrdinates of the positions of pulse sources were determined by measuring the times at which the noise pulses arrived at each of the spaced receivers. Lightning channels were mapped in space and time by locating a large number of radio noise sources for each flash, whose shape and position then became apparent despite the presence of intervening hydrometeors. A center frequency of 253 MHz was chosen for the receivers and their bandwidths were wide enough for time differences to be measured with rms errors of 140 ns. Hence two of the three spatial coordinates of a particular source could be determined to an accuracy of 25 m rms and the vertical coordinate could be found to an accuracy which was typically 140 m rms, but the actual height accuracy depended on source position. Some factors that affect the design of the receivers are discussed and case studies of five cloud flashes are presented. Cloud flashes could be classified into two types according to the rates at which they emitted pulses in the VHF part of the radio spectrum. One class radiated pulses at rates that approximated 103 pulses per second when received in bandwidths of 10 MHz. These pulses were often nearly rectangular in form, lasted approximately l ms on average, and occurred in synchronism with pulses received at HF and at UHF. The second class emitted much shorter pulses (median duration of 0.2 to 0.4 ms were measured) at higher rates typically 105 pulses per second, and pulses were generally not in synchronism with those received at other radio frequencies. Diameters of the channels occupied by radio sources varied from 100 m to several hundred meters, and were enlarged by the extents of the sources themselves. It was possible to measure the principal extents of many individual sources active during low-pulse-frequency cloud flashes. Average sizes near 300 m were measured for low-pulse-frequency flashes, and sizes near 60 m were estimated for sources active during high-pulse-frequency flashes. It was found that pulses originated in regions near streamer tips, and that pulses were associated with initial ionization. First streamer progressed at speeds that ranged from 0.9¿105 m/s to 2.1¿105 m/s except for one extensive, positive flash that moved a 5¿105 m/s for the first few kilometers. We define a positive flash as being one which conducts excess positive charge in the same direction as that in which the streamer progresses. Subsequent discharges along paths that had been ionized previously seldom radiated much noise but noise was radiated by channels that were several ten of milliseconds old, and then we measured streamer speeds that were at least an order of magnitude higher than first streamer speeds. Relatively few clouds flashes were found to have been oriented vertically. Most were horizontal and often consisted of several streamers that extended from a common origin. Measurements of electric field change were used to estimate quantities of charge involved. These estimates were made by adopting various models for the distribution of charge along the know paths, and the quantities were not greatly dependent on the form of the distribution. Most estimates yielded line densities near 10-3 C m-1. Only one of the cloud discharges had been a positive flash. Trains of band-limited noise which lasted for times that ranged approximately from 10 μs to about 1.5 ms were also emitted by lightning flashes of all kinds, but in cloud flashes this type of noise, which was distinctly different from the pulsed emisson, occurred more frequently and with longer durations during the J-type portions or final stages of cloud flashes. It accompanied streamers which progressed at speeds near 107 m/s, often accompanied K changes, which were usually delayed after the start of this noise by tens of microseconds, and the streamers that caused the noise were positive streamers, with the single exception of one train which was found to have been due to a negative streamer that formed near the origin of an extensive, positive cloud flash. There was evidence that events that produced this noise sometimes triggered dart leaders and recoil streamers and most sources of this band-limited noise were located near the origins of the flashes, and streamers which caused the noise served also to discharge the origins but not usually by way of the paths formed by first streamers. The principal results of this work are listed at the end of the paper.