Functional magnetic resonance imaging (fMRI) based on blood oxygenation level-dependent (BOLD) contrast allows non-invasive examination of brain activity and is widely used in the neuroimaging field. The BOLD contrast mechanism reflects hemodynamic changes resulting from a complex interplay of blood flow, blood volume, and oxygen consumption. Heart rate (HR) variations are the most intriguing and less understood physiological processes affecting the BOLD signal, as they are the result of a wide variety of interacting factors. The use of the response function that best models HR-induced signal changes, called cardiac response function (CRF), is an effective method to reduce HR noise in fMRI. However, current models of physiological noise correction based on CRF, i.e. canonical and individual, either do not take into account variations in HR between subjects, and are thus inadequate for cohorts with varying HR, or require time-consuming quality control of individual physiological recordings and derived CRFs. By analyzing a large cohort of healthy individuals, the results presented in this thesis show that different HRs influence the BOLD signal and their corresponding spectra differently. A further finding is that HR plays an essential role in determining the shape of the CRF. Slower HRs produce a smoothed CRF with a single well-defined maximum, while faster HRs cause a second maximum. Taking advantage of this dependence of the CRF on HR, a novel method is proposed to model HR-induced fluctuations in the BOLD signal more accurately than current approaches of physiological noise correction. This method, called HR-based CRF, consists of two CRFs: one for HRs below 68 bpm and one for HRs above this value. HR-based CRFs can be directly applied to the fMRI data without the time-consuming task of deriving a CRF for each subject while accounting for inter-subject variability in HR response.