In this work, we study the structure of road traffic with the help of fractal and multifractal tools. Using classical models of traffic that lead to a Burgers' equation and recent results on the solutions of this equation when the initial conditions are scaling, we predict that, under some circumstances, the traffic can possess a multifractal structure similar to those of multiplicative processes. We then verify this behavior on six minute data of traffic flows. The high sampling rate allows to evidence the highly irregular nature of the flows and to quantify this irregularity using the classical tools of the multifractal theory, namely the (q, A(q)) and the (A, f(A)) curves. These characterizations in turn permit to classify the complex traffic data, with some application to short-term prediction.