Most physiological time series have self-similar properties which reflect the functioning of physiological control mechanisms. Self-similarity is usually assessed by detrended fluctuation analysis (DFA) assuming that mono- or bi-fractal models generate the self-similar dynamics. Our group recently proposed a new DFA approach describing self-similarity as a continuous temporal spectrum of coefficients, thus not assuming that "lumped-parameter" fractal models generate the data. This paper reviews the rationale for calculating a spectrum of DFA coefficients and applies this method on datasets of signals whose self-similarity has been extensively studied in the past. The first dataset consists of six electroencephalographic (EEG) derivations collected in a healthy volunteer. The second dataset consists of cardiac intervals and diastolic blood pressures recorded in 60 volunteers at different levels of cardiac sympatho/vagal balance. Results reveal the limits of the traditional "lumped-parameter" approach, and provide information on the role of autonomic outflows in determining cardiovascular self-similarity.