We consider a distributed communication network with many terminals which are distributed in space and wish to communicate with each other using a common radio channel. Choosing the transmission range in such a network involves the following tradeoff: a long range enables messages to reach their destinations in a few hops, but increases the amount of traffic competing for the channel at every point. We give a simple model for the per-hop delay in random access networks, analyze this tradeoff, and give the optimal transmission range. When choosing this optimal range, as a function of specified traffic and delay parameters, networks demonstrate an important self-adjusting capability. This capability to adjust to traffic makes heavily loaded networks far better than centralized systems (in which all messages must reach one common destination). Dividing a terminal population into power groups can improve any random access system, especially when the traffic is split between groups in an appropriate way, which we demonstrate. But since networks are hurt by destructive interference less than centralized systems, it is harder to improve them. Using power groups can significantly improve centralized systems, but will lead to a smaller relative improvement in networks. Decomposing the system into a hierarchy of ALOHA levels, with only a small population contending at the top level, can improve centralized systems but does not improve networks.