The paper presents a survey and comparison of techniques for automatic facial action classification. The Facial Action Coding System (FACS) is an objective method for quantifying facial movements, widely used in behavioral science. The study compares various techniques for automatically recognizing facial actions, including optical flow analysis, principal component analysis (PCA), independent component analysis (ICA), and Gabor wavelet representations. These methods were evaluated against naive and expert human subjects. The best performance was achieved using Gabor wavelet and ICA representations, both achieving 96% accuracy in classifying 12 facial actions. The results highlight the importance of local filters, high spatial frequencies, and statistical independence in facial action classification. The study also explores different image representations for facial expression recognition, including holistic and local spatial analysis. The paper discusses the advantages and limitations of various techniques, emphasizing the need for accurate and reliable facial action classification in behavioral science and human-computer interaction. The findings suggest that image-based approaches, which directly learn facial action classes from image sequences, can provide more accurate and detailed information than physical models. The study concludes that the best performance in facial action classification is achieved using local filters, high spatial frequencies, and statistical independence.The paper presents a survey and comparison of techniques for automatic facial action classification. The Facial Action Coding System (FACS) is an objective method for quantifying facial movements, widely used in behavioral science. The study compares various techniques for automatically recognizing facial actions, including optical flow analysis, principal component analysis (PCA), independent component analysis (ICA), and Gabor wavelet representations. These methods were evaluated against naive and expert human subjects. The best performance was achieved using Gabor wavelet and ICA representations, both achieving 96% accuracy in classifying 12 facial actions. The results highlight the importance of local filters, high spatial frequencies, and statistical independence in facial action classification. The study also explores different image representations for facial expression recognition, including holistic and local spatial analysis. The paper discusses the advantages and limitations of various techniques, emphasizing the need for accurate and reliable facial action classification in behavioral science and human-computer interaction. The findings suggest that image-based approaches, which directly learn facial action classes from image sequences, can provide more accurate and detailed information than physical models. The study concludes that the best performance in facial action classification is achieved using local filters, high spatial frequencies, and statistical independence.