The hazardous potential of hydrogen/air mixtures has intensively been studied assuming a perfect mixture of fuel and oxidant. However, comprehensive risk assessment studies have shown that an inhomogeneous mixture with a vertical concentration gradient is much more likely to be generated in a real accident scenario. From a safety point of view, an open question is whether established criteria such as the 7 lambda criterion for the determination of the deflagration-to-detonation transition (DDT) limits can be applied to inhomogeneous mixtures as well. For the experimental investigation of DDT in such mixtures an injection mechanism has been developed that produces vertical fuel concentration gradients inside a horizontal channel with large aspect ratio. The channel is equipped with obstacles to enhance flame acceleration. Photodiodes and pressure transducers measure flame and shock arrival times. Schlieren measurements are conducted to track gas mixing behind the obstacles and to track the formation of the detonation wave front. In the experimental study, the fuel content, the strength of concentration gradient, the blockage ratio, and the spacing of the obstacles inside the channel are varied. The maximum flame velocity observed in each experiment is compared with the velocities obtained from one-dimensional theory. DDT leads to a strong velocity rise from the sound speed of the combustion products to the Chapman-Jouguet velocity of the mixture and generates a considerable rise in maximum pressure. It is found that in homogeneous mixtures, the 7 lambda criterion often predicts DDT reasonably well. However, concentration gradients shift the point of DDT either to considerably higher or to lower fuel concentrations. As multidimensional effects occur, which depend on the specific configuration of the obstacles with respect to the mixture gradient, the simple 7 lambda criterion is no longer valid.