Retrieval schemes are developed based on neural network techniques to derive atmospheric temperature and humidity profiles from the NOAA-15 and NOAA-16 Advanced Microwave Sounding Unit-A (AMSU-A) and Unit-B (AMSU-B) measurements. The neural network training sets are built based on the pole-to-pole global AMSU-A and AMSU-B measurements and the corresponding analysis data generated at the National Center for Environmental Prediction. For the temperature retrieval, all 15 channels of AMSU-A data are used to maximize the vertical resolution of the measurement. The preliminary study based on five days of global AMSU-A data from a winter season has yielded good retrieval results for the temperature profiles from the near-surface levels to 10 hPa pressure level. Further study is underway that includes the global AMSU-A data from other seasons. For the humidity profile retrieval, the scheme is based on the 15-channel AMSU-A measurement and the 5-channel AMSU-B measurement. These channels either contribute directly to water vapor measuring, or set constraints to the retrieval pattern. The retrieval of specific humidity profile shows the root mean square deviations of 1.1 g/kg at the surface, between 0.8 and 1 g/kg in the lower atmosphere (700 - 1000 hPa), and much smaller values toward the middle and upper atmosphere.