Understanding the cloud response to external forcing is a major challenge for climate science. This crucial goal is complicated by inter-model differences in simulating present and future cloud cover and by observational uncertainty. Here, we perform the first formal detection and attribution study of cloud changes over the satellite era. We present CMIP5 model-derived fingerprints of externally forced changes to three cloud properties: the latitudes at which the zonally averaged total cloud fraction (CLT) is maximized or minimized, the zonal average CLT at these latitudes, and the height of high clouds at these latitudes. By considering simultaneous changes in all three properties, we define a coherent multivariate fingerprint of cloud response to external forcing and use CMIP5 models to calculate the average time to detect these changes. We find that given perfect satellite cloud observations beginning in 1983, the models indicate that a detectable multivariate signal should have already emerged. We then search for signals of external forcing in two observational datasets: ISCCP and PATMOS-x. The datasets are both found to show a poleward migration of the zonal CLT pattern that is incompatible with forced CMIP5 models. Nevertheless, a detectable multivariate signal is predicted by models over the PATMOS-x time period and is indeed present in the dataset. Despite persistent observational uncertainties, our results present a strong case for continued efforts to improve these existing satellite observations, in addition to planning for new missions.