Across neuroscience research, clinical diagnostics, and engineering applications in pain evaluation and treatment, there is a need for an objective measure of pain experience and detection when it occurs. This detector should be reliable in real-world settings using easily accessible, non-invasive data sources. We present a simple yet robust paradigm for decoding pain using neural and physiological data including electroencephalography (EEG), pulse, and skin conductance (GSR) measurements. The present study uses multivariate classification to distinguish painful events from non-painful multimodal sensory stimuli. To classify the pain response and detect relevant data attributes, we employed a sparse logistic regression (SLR) machine learning protocol with automatic feature selection. EEG input consisted of time-frequency changes under trial conditions, and physiological data included fluctuations and spikes in pulse and skin conductance. Classification averaged 70% accuracy and selected between 5 and 15 features. In our experiment, pain was induced by cold stimulation which became noxious with prolonged exposure. Due to the long, ramp-and-hold nature of the stimulus, along with individual variability in sensitivity to pain, we did not observe specific rapid evoked responses or time-locked events common across participants. However, this format more closely resembles the experience of pain conditions requiring intervention which could be facilitated by a decoding system. The results illustrate the feasibility of developing a wireless pain detection system and give insight to important temporal, spectral, and spatial EEG events and physiological indicators of pain states. Success of the classifier protocol using these parameters could lead to the creation of a closed-loop system for decoding and intervention which can be applied in engineering and medical contexts.