Magnetic storms are mainly caused by the occurrence of intense southward magnetic fields in the interplanetary medium. These fields can be formed directly either by ejection of magnetic structures from the Sun or by stream interaction processes during solar wind propagation. In the present study we examine 30 years of satellite measurement of the solar wind during magnetic storms, with the aim of estimating the relative importance of these two processes. We use the solar wind proton temperature relative to the temperature expected from the empirical relation to the solar wind speed Tp/Texp, together with the speed gradient, and the interplanetary magnetic field azimuth in the ecliptic, in order to distinguish between the two processes statistically. We find that compression due to stream interaction is at least as important as the direct effect of ejection of intense fields, and probably more so. Only around 20--25% of major and large storm hours and 10--15% of medium and small storm hours are directly associated with a solar wind meeting the criteria Tp/Texp<0.5, a criteria which previously has been found to be a good indicator of ejected material. More than 50% of the storm hours are caused by a solar wind having an enhanced proton temperature relative to what is expected on the basis of the observed speed and having clear traces of gardenhose orientation of the magnetic field in the ecliptic plane. ¿ 2001 American Geophysical Union