This paper summarizes the results of several projects, which were aiming at comparing different microscopic traffic flow models to reality and with each other. Surprisingly, it was not possible to calibrate model with an error smaller than 15 % (relative root mean square error, RRME) against real data. Simple models basically yield the same results as more complicated ones, where the number of parameters has been used as a measure of complexity of a model. Before conclusions can be drawn out of these results, it must be understood into more detail the sources of noise in traffic flow. This will be undertaken by running models with NGSIM trajectory data