Respiratory sinus arrhythmia (RSA) has been widely used as a measure of the cardiac vagal control in response to stress. However, RSA seems not to be a generalized indicator because of its dependency on respiratory parameter and individual variations of RSA amplitude (A(RSA)). We hypothesized that phase-lag variations between RSA and respiration may serve as a normalized index of the degree of mental stress. Twenty healthy volunteers performed mental arithmetic task (ART) after 5 min of resting control followed by 5 min of recovery. Breathing pattern, beat-to-beat R-R intervals, and blood pressure (BP) were determined using inductance plethysmography, electrocardiography, and a Finapres device, respectively. The analytic signals of breathing and RSA were obtained by Hilbert transform and the degree of phase synchronization (lambda) was quantified. With the use of spectral analysis, heart rate variability (HRV) was estimated for the low-frequency (LF) and high-frequency (HF) bands. A steady-state 3-min resting period (REST), the first 3 min (ART1), and the last 3 min (ART2) of the ART period (ranged from 6- to 19 min) and the last 3 min of the recovery period (RCV) were analyzed separately. Heart rate, systolic BP, and breathing frequency (f(R)) increased and lambda, A(RSA), and HF power decreased from REST to ART (P < 0.01). The lambda was correlated with normalized A(RSA) and the HF power. The decrease in lambda could not be explained solely by the increase in f(R). We conclude that mental stress exerts an influence on RSA oscillations, inducing incoherent phase lag with respect to breathing, in addition to a decrease in RSA.