Detecting pain in infants is of vital importance in healthcare. The currently applied pain scaling methods are subjective and it is not possible to apply them continuously because they requires the aid of trained healthcare professionals. This work proposes and investigates two different systems for automated continuous facial analysis for stress and pain detection in infants. Both systems are based on the same face detection approach. The first system uses an Active Appearance Model (AAM) and a three-class SVM classifier. The second system detects three Regions Of Interest (ROI), aiming at detecting the presence of the brow bulge, eye squeeze and the nasolabial furrow. For this system, the resulting pain/stress level is detected with an accuracy of 67%. It follows the PIPP pain scale form and is able to detect the facial regions, even with occlusions like feeding tubes. The first (AAM-based) system is not able to handle occlusions but has an accuracy of 92%, classifying the facial expressions into comfort, discomfort and the primal face of pain (PFP).