Heart rate variability (HRV) has become an important parameter for the assessment of autonomic function in many areas of medicine. In particular, respiratory sinus arrhythmia measured during the deep breathing test (DBT) is often used. Results are usually expressed in common time-domain parameters. A "most preferred measure" has not yet been identified. We investigated the sensitivity of the DBT to the following anomalies: in-test variance; shifts of mean heart rate; premature ventricular contractions; and breathing rate deviations. Frequency and magnitude of the anomalies were determined in a set of real DBTs (n=514) and transferred to computer simulations to mimic realistic conditions. The sensitivity of standard deviation, mean circular resultant (MCR), root mean square of successive differences (RMSSD), and four types of expiration-inspiration (E-I) difference were quantified statistically. Median-based E-I differences, E-I ratio, and MCR were most resistant to the anomalies. E-I difference derived by median values should be used preferentially, providing the highest precision and independence from heart rate.