The aim of this work was to evaluate the performance of a novel non-invasive continuous glucose-monitoring system based on impedance spectroscopy (IS) in patients with diabetes. Ten patients with type 1 diabetes (mean+/-S.D., age 28+/-8 years, BMI 24.2+/-3.2 kg/m(2) and HbA(1C) 7.3+/-1.6%) and five with type 2 diabetes (age 61+/-8 years, BMI 27.5+/-3.2 kg/m(2) and HbA(1C) 8.3+/-1.8%) took part in this study, which comprised a glucose clamp experiment followed by a 7-day outpatient evaluation. The measurements obtained by the NI-CGMD and the reference blood glucose-measuring techniques were evaluated using retrospective data evaluation procedures. Under less controlled outpatient conditions a correlation coefficient of r=0.640 and a standard error of prediction (SEP) of 45 mg dl(-1) with a total of 590 paired glucose measurements was found (versus r=0.926 and a SEP of 26 mg dl(-1) under controlled conditions). Clark error grid analyses (EGA) showed 56% of all values in zone A, 37% in B and 7% in C-E. In conclusion, these results indicate that IS in the used technical setting allows retrospective, continuous and truly non-invasive glucose monitoring under defined conditions for patients with diabetes. Technical advances and developments are needed to expand on this concept to bring the results from the outpatient study closer to those in the experimental section of the study. Further studies will not only help to evaluate the performance and limitations of using such a technique for non non-invasive glucose monitoring but also help to verify technical extensions towards a IS-based concept that offers improved performance under real life operating conditions.