We present an original end-to-end available bandwidth measurement method, called SMART (Statistics Measurement for Avail-bw by Random Train). It resolves some of the problems common for many types of existing probing methods, e.g. the long latency and large probe traffic. Contrary to traditional estimates of available bandwidth, SMART is not a methodology based on packet dispersion in packet pair or packet train, but a completely new methodology in the light of probability and statistics. The fundamental idea is to send very small packets at random moment and calculate the proportion of minimal delay in total test samples. To reach this purpose, we redefine the available bandwidth based on probability and statistics. We have evaluated our method in controlled and reproducible environment using NS2, and the simulations show our method is accurate, efficient, quick and non-intrusive.