Semiconductor manufacturers are forced by market demand to continually deliver lower cost and faster devices. This results in complex industrial processes that, with continuous evolution, aim to improve quality and reduce costs. Plasma etching processes have been identified as a critical part of the production of semiconductor devices. It is therefore important to have good control over plasma etching but this is a challenging task due to the complex physics involved. Optical Emission Spectroscopy (OES) measurements can be collected non-intrusively during wafer processing and are being used more and more in semiconductor manufacturing as they provide real time plasma chemical information. However, the use of OES measurements is challenging due to its complexity, high dimension and the presence of many redundant variables. The development of advanced analysis algorithms for virtual metrology, anomaly detection and variables selection is fundamental in order to effectively use OES measurements in a production process. This thesis focuses on computational intelligence techniques for OES data analysis in semiconductor manufacturing presenting both theoretical results and industrial application studies. To begin with, a spectrum alignment algorithm is developed to align OES measurements from different sensors. Then supervised variables selection algorithms are developed. These are defined as improved versions of the LASSO estimator with the view to selecting a more stable set of variables and better prediction performance in virtual metrology applications. After this, the focus of the thesis moves to the unsupervised variables selection problem. The Forward Selection Component Analysis (FSCA) algorithm is improved with the introduction of computationally efficient implementations and different refinement procedures. Nonlinear extensions of FSCA are also proposed. Finally, the fundamental topic of anomaly detection is investigated and an unsupervised variables selection algorithm tailored to anomaly detection is developed. In addition, it is shown how OES data can be effectively used for semi-supervised anomaly detection in a semiconductor manufacturing process. The developed algorithms open up opportunities for the effective use of OES data for advanced process control. All the developed methodologies require minimal user intervention and provide easy to interpret models. This makes them practical for engineers to use during production for process monitoring and for in-line detection and diagnosis of process issues, thereby resulting in an overall improvement in production performance.