Imaging of the Amazon deep-sea fan with long-range side-scan sonar (GLORIA) has, for the first time, revealed the anatomy, trends, and growth pattern of distributary channels on this fan. Only one channel-levee system was active at any given time and extended from the Amazon Submarine Canyon downslope onto the lower fan (> 4,200 m). Formation of new channel-levee systems occurred when a currently active channel-levee system was cut off and abandoned through avulsion, and a new channel-levee system was established nearby. Through time, successive channel-levee formation and abandonment built two broad levee complexes consisting of groups of overlapping, coalescing segments of channel-levee systems across the present fan surface. These, plus older, now buried levee compl xes, indicate that fan growth is radially outward and downslope through development of successive levee complexes.

The most striking characteristic of the distributary channels is their intricate, often recurving, meanders with sinuosities of up to 2.5. Cutoffs and abandoned meander loops indicate that the channels migrate laterally through time. Channel bifurcation results predominantly from avulsion when flows breach a channel levee, thereby abandoning the present channel and establishing a new channel-levee segment nearby. No clear evidence of channel branching (i.e., division of a single channel into two active segments) or braiding was observed.

The formation, maintenance, and modification of such extensive and intricate meander systems presumably requires large volumes of relatively continuous turbidity flow through channels for relatively long time periods. "Classic" sporadic turbidity-current events are apparently too infrequent to form the meandering channels. However, the bulk of sediments forming the channel-levee systems consists of fine terrigenous (hemipelagic) sediments, which were deposited relatively continuously throughout entire intervals (~ 100,000 years) of glacio-eustatic sea level lowering (e.g., Wisconsin). Although defining precisely what "continuous" means is not yet possible in terms of frequency of individual flow events (per year, decade, etc), this type of more continuous, high-volume flow is probably an important factor that promotes channel meandering. Another important factor that appears to cause channel meandering is changes in the slope of the fan surface (i.e., valley slope) encountered by a channel-levee system as it builds across the fan. The channel is apparently able to maintain a preferred gradient or slope along its thalweg, despite an increase in the slope of the fan surface, simply by meandering.