||In the past decade, there has been a growing interest for the development of olfactory machines and Electronic Nose (EN) systems in order to fulfill a vari-ety of real-life applications. A number of applications have recently emerged in the area of safety (monitoring leakage of combustible gases), environmental ap-plications (air quality and pollution control), medical applications and mobile robot navigation. New development of microelectronic gas sensors has recently enabled integrated low power EN system. Unfortunately the gas sensors do suffer a number of serious shortcomings such as non-selectivity, non-linearity, drift and slow response. To overcome some of these problems, an EN system typically includes an array of sensors followed by signal processing stages. To date, most of EN systems reported in the literature rely on software imple-mentation of the processing stages. Within this background, the contributions of this thesis are three-fold: At the sensor level, a sensor characterization platform was developed and a very comprehensive experimental data set was collected for gas detection using tin oxide gas sensors. The advantage and shortcomings of the sensor were also identified. Redundancy analysis for gas detection based on the tin oxide gas sensor array was performed. At the algorithmic level, a wide range of pre-processing techniques as well as pattern recognition algorithms (KNN, MLP, RBF, GMM, PPCA and SVM) were compared for the applications at hand. Based on these algorithms, a committee machine, which combines different classifiers was built in order to build a more accurate classifier. The commit-tee machine relies on a novel voting and weighting functions which permits to build a robust classifier. At the hardware level, hardware-friendly architectures allowing to reduce the hardware resources for implementing the proposed clas-sifier was proposed without affecting the classification performance. The effect of both hardware implementation and quantization errors on the classification performance was thoroughly investigated. It was found that the committee machine was more robust than other classifiers at lower precision. A hardware friendly digital VLSI implementation of GMM classifier using a novel pipelin-ing strategy and piecewise linear approximation was implemented in 0.25μm CMOS process. Results showed that the classification of 100 gas patterns can be performed in 57μs with a classification accuracy of 92.5%. The overall gas identification system was implemented using dynamically reconfigurable FPGA. The system can be dynamically configured to accommodate different stages of the processing at different times allowing to efficiently share limited hardware resources of the FPGA. Test results demonstrate the effectiveness of the committee machine for gas sensor applications.