<p>Functional magnetic resonance imaging (fMRI) is a valuable neuroimaging tool for studying brain function and connectivity. However, the blood oxygen level dependent (BOLD) signal used in fMRI is affected by various physiological factors, such as cardiac and respiratory activity, which can influence functional connectivity patterns. As such, physiological noise correction is a crucial preprocessing step in fMRI data analysis. When concurrent physiological recordings are available, researchers often generate nuisance regressors to account for the effect of heart rate and respiratory variations by convolving physiological response functions (PRF) with the corresponding physiological signals. However, it has been suggested that the PRF characteristics may vary across subjects and different regions of the brain, as well as across scans of the same subject. To explore this variability, we examine the performance of several different PRF models, in terms of BOLD variance explained, using resting-state fMRI data from the Human Connectome Project (<i>N</i> = 100). We examined both one-input (heart rate or respiration) and two-input (heart rate and respiration) PRF models and show that allowing PRFs to vary across subjects and brain regions generally improves PRF model performance. For one-input models, the improvement in model performance gained by allowing spatial variability was most prominent for respiration, particularly for a subset of the subjects (about a third) examined. Subject-specific or regional variability in cardiac response only enhanced performance when using two-input models. Overall, our results highlight the importance of considering spatial and subject-specific variability in PRFs when analyzing fMRI data, particularly regarding respiratory-related fluctuations.</p>