Accuracy of the atmospheric profiles of temperature and humidity, retrieved from infrared sounder observations using physical retrieval algorithm, depend primarily on the quality of the first guess profiles. In the past, forecasts from the numerical weather prediction models were extensively used as the first guess. During past few years, the first guess for physical retrieval is being estimated using regression techniques from sounder observations. In the present study, a new non-linear technique has been described to improve the first guess using simulated infrared brightness temperatures for GOES-12 Sounder channels. The present technique uses fuzzy logic and data clustering to establish a relationship between simulated sounder observations and atmospheric profiles. This relationship is further strengthened using Adaptive Neuro-Fuzzy Inference System (ANFIS) by fine-tuning the existing fuzzy rule base. The results of ANFIS retrieval have been compared with the non-linear (polynomial) regression retrieval. It has been found that ANFIS is more robust and shows remarkable improvement as it reduces RMS error by 20% in humidity profiles retrieval compared to the non-linear regression technique. In addition, it has been shown that the ANFIS technique has an added advantage of its global application without any need for training data classification that is required in the regression techniques.