The paper deals with methods of processing ECG and respiration signals which aim at detecting parameters whose values may be correlated to normal and diabetic subjects with or without cardiovascular autonomic neuropathy (CAN). Beat-to-beat R-R duration values of the ECG and discrete series of respiration are obtained from original signals using a recognition algorithm. Power spectrum analysis (autospectra, cross-spectra and coherence via autoregressive modelling) is carried out on segments of about 200 consecutive cardiac cycles. Spectral parameters of the R-R variability signal are obtained as follows: total power, powers of low-frequency (LF) and high-frequency (HF) components, power of the signal which is (or is not) coherent with respiration, in absolute or in percentage values. The experimental protocol considers 40 diabetic patients (21 of whom have diabetic neuropathy) and 14 normals in three different conditions: resting, standing and controlled respiration. The developed spectral parameters seem sensitive enough to differentiate between normal and pathological subjects. These parameters may constitute a quantitative means to be added to the classical diabetic tests for the diagnosis of cardiovascular autonomic neuropathy.