Continuous Glucose Monitoring(CGM) devices allow sensor augmented pump therapy for type-1 diabetic patients. They are also useful for physicians to analyze nycthemeral glycemic profile and glucose variability. Noise, deviations, loss of sensitivity and spikes are common sources of error in CGM readings. An off-line wavelet algorithm has been developed to denoise the CGM signal. A typical CGM noise was added to a simulated blood glucose concentration.The objective is to maximize the signal to noise ratio. The Haar wavelet basis has been selected. The chosen scale J of decomposition gives a sequence of scale coefficients which contains an approximation of the signal at the resolution scale J, and a family of detail coefficients dj, 1≤ j ≤ J. Optimal scale of decomposition and threshold levels are computed. CGM data from 16 type-1 diabetic patients were used for this study. The optimal threshold levels were used to denoise the CGM signal off-line. The results showed that noise, deviations, loss of sensitivity and spikes were removed from the signal ; likely pikes and nadirs were not damped ; the algorithm dealt with loss of signal. This algorithm can provide a pre-processing tool for model fitting and allows an accurate reading of CGM to facilitate analysis by the physicians.