Detrended fluctuation analysis (DFA) is becoming a widely used technique for exploring the structure of correlations in heart rate variability (HRV) data. This method provides a scaling or fractal exponent alpha(x) derived from the behaviour of the root-mean-square fluctuations along different time scales n. Rather than just finding a single exponent, covering either short (alpha(1)) or long range (alpha(2)), we recently suggested tracking the local evolution of alpha(x), as in this way scaling patterns (SP), which seem to provide more detailed characterizations of HRV data, are revealed. Here, we evaluate such potential advantage by classifying long-term data from 51 subjects in normal sinus rhythm and 29 congestive heart failure patients. Using the SP we achieved a significantly better classification of these data than using alpha(x), or the statistic pNN20, thereby confirming that the SP provide a useful assessment of the correlation structure in HRV data.