Most stars form as part of a star cluster. The most massive clusters in the Milky Way exist in two groups - loose and compact clusters - with significantly different sizes at the end of the star formation process. After their formation both types of clusters expand up to a factor 10-20 within the first 20 Myr. Gas expulsion at the end of the star formation process is usually regarded as only possible process that can lead to such an expansion. We investigate the effect of gas expulsion by a direct comparison between numerical models and observed clusters concentrating on clusters with masses >103M(sun). For these clusters the initial conditions before gas expulsion, the characteristic cluster development, its dependence on cluster mass, and the star formation efficiency
(SFE) are investigated. We perform N-body simulations of the cluster expansion process after gas expulsion and compare the results with observations. We find that the expansion processes of the observed loose and compact massive clusters are driven by completely different physical processes. As expected the expansion of loose massive clusters is largely driven by the gas loss due to the low SFE of ~30%. One new revelation is that all the observed massive clusters of this group seem to have a very similar size of 1-3 pc at the onset of expansion. It is demonstrated that compact clusters have a much higher effective SFE of 60-70% and are as a result much less affected by gas expulsion. Their expansion is mainly driven by stellar ejections caused by interactions between the cluster members. The reason why ejections are so efficient in driving cluster expansion is that they occur dominantly from the cluster centre and over an extended period of time. Thus during the first 10 Myr the internal dynamics of loose and compact clusters differ fundamentally.