Early detection of oil spill events is the key to environmental protection and disaster management. Current technology lacks the sensitivity and specificity in detecting the early onset of a small-scale oil spill event. Based on an infrared oil-water contrast model recently developed, we propose a novel nonscanning computational infrared sensor that has the potential to achieve unprecedented detection sensitivity. Such a system can be very low-cost and robust for automated outdoor operations, leading to massive offshore deployment. Taking advantage of the characteristic oil thickness multispectral signatures, we have streamlined an algorithm that incorporates 3D image reconstruction and classification in a single inversion step capitalizing on the benefits of L1 minimization.