This study assesses the detectability of external influences in changes of precipitation extremes in the twentieth century, which is explored through a perfect model analysis with an ensemble of coupled global climate model (GCM) simulations. Three indices of precipitation extremes are defined from the generalized extreme value (GEV) distribution: the 20-year return value (P20), the median (Pm), and the cumulative probability density as a probability-based index (PI). Time variations of area-averages of these three extreme indices are analyzed over different spatial domains from the globe to continental regions. Treating all forcing simulations (ALL; natural plus anthropogenic) of the twentieth century as observations and using a preindustrial control run (CTL) to estimate the internal variability, the amplitudes of response patterns to anthropogenic (ANT), natural (NAT), greenhouse-gases (GHG), and sulfate aerosols (SUL) forcings are estimated using a Bayesian decision method. Results show that there are decisively detectable ANT signals in global, hemispheric, and zonal band areas. When only land is considered, the global and hemispheric detection results are unchanged, but detectable ANT signals in the zonal bands are limited to low latitudes. The ANT signals are also detectable in the Pm and PI but not in P20 at continental scales over Asia, South America, Africa, and Australia. This indicates that indices located near the center of the GEV distribution (Pm and PI) may give better signal-to-noise ratio than indices representing the tail of the distribution (P20). GHG and NAT signals are also detectable, but less robustly for more limited extreme indices and regions. These results are largely insensitive when model data are masked to mimic the availability of the observed data. An imperfect model analysis in which fingerprints are obtained from simulations with a different GCM suggests that ANT is robustly detectable only at global and hemispheric scales, with high uncertainty in the zonal and continental results.