Spinal cord functional imaging allows assessment of activity in primary synaptic connections made by sensory neurons relaying information about the state of the body. However, reported human data based on gradient-echo techniques have been largely inconsistent, with no clear patterns of activation emerging. One reason for this variability is the influence of physiological noise, which is typically not corrected for. By acquiring single-slice resting data from the spinal cord with a conventional gradient-echo EPI pulse sequence at TR=200 ms (critically sampled) and TR=3 s (under-sampled), we have characterised various sources of physiological noise. In 8 healthy subjects, the presence of physiologically dependent signal was explored using probabilistic independent component analysis (PICA). Based on the insights provided by PICA, we defined a new physiological noise model (PNM) based on retrospective image correction (RETROICOR), which uses independent physiological measurements taken from the subject to model sources of noise. Statistical significance of individual components included in the PNM was assessed by F-tests, which demonstrated that the optimal PNM included cardiac, respiratory, interaction and low-frequency regressors. In a group of 10 healthy subjects, activation data were acquired from the cervical spinal region (T1 to C5) during painful thermal stimulation of the right and left hands. The improvement obtained when using a PNM in estimating spinal cord activation was reflected in a reduction of false-positive activation (active voxels in the CSF space surrounding the cord), when compared to conventional GLM modelling without a PNM.