We report on a general and automatic data-driven background distribution\nshape estimation method using neural autoregressive flows (NAF), which is one\nof the deep generative learning methods. Data-driven background estimation is\nindispensable for many analyses involving complicated final states where\nreliable predictions are not available. NAF allow us to construct general\nbijective transformations that operate on multidimensional space, out of finite\nnumber of invertible one-dimensional functions. Given its simplicity and\nuniversality, it is well suited to the application in the data-driven\nbackground estimation, since data-driven estimations can be expressed as\ntransformations. In a data-driven background estimation, the goal is to derive\nappropriate transformations and apply extrapolated transformations to the\nregion of interest. In the ABCDnn method, we can have the NAF learn the\ntransformations' dependence on control variables by having multiple control\nregions. We demonstrate that the prediction through ABCDnn method is similar to\noptimal case, while having smaller statistical uncertainty.\n