Solar tornadoes are thought to be produced by rotating magnetic field structures which extend from the upper convection zone and the photosphere to the corona of the Sun. Recent studies show that rotating motions are an integral part of atmospheric dynamics and occur on a large range of spatial scales. A systematic statistical study of magnetic tornadoes is a necessary next step towards understanding their formation and their role for the mass and energy transport in the solar atmosphere. We have developed a new automatic detection method for solar tornadoes. Unlike the previous studies that relied on visual inspections, our new method combines local correlation tracking (LCT) and line integral convolution (LIC) techniques. For disk centre observations, the LCT gives us information on the horizontal velocity field which is indispensable for searching for solar tornados. The LIC helps us to visualise the velocity maps, which are given by the LCT, and finally to identify swirls (i.e., the observational signatures of tornadoes). We have tested several detection algorithms for a variety of spatial/temporal scales based on a systematic set of numerical model atmospheres created with CO5BOLD. We present first results from an analysis of these simulations and observations, which strongly imply that magnetic tornadoes are most frequent in quiet Sun/enhanced network regions, where they might contribute significantly to the energy transport into the upper atmosphere.