The prediction of network traffic flow is a problem of great significance in the research work of resource allocation and congestion control. Based on this accurate prediction, the scheme of resource allocation and control can easily adapt to dynamic variations of incoming traffic flow. So the goal of optimal network performance is achieved. There are many sorts of traffic flow of self-similarity characteristics in high-speed network, and some research work has showed this self-similarity keeps in close contact with the attractor of chaos system. A rate prediction of self-similar traffic sources in high-speed network is proposed as well as the maximum of predictable time by applying the technology of phase space reconstruction about chaotic time series. This method has a simple prediction mode, and the result of simulations indicates it also has highly accurate results.