The baseline correction is an, extremely important spectral preprocessing step and can significantly improve the accuracy of the subsequent spectral analysis algorithm. At present most of the baseline correction algorithms are manual and semi-automated. The manual baseline correction depends on the user experience and its accuracy is greatly affected by the subjective factor. The semi-automated baseline correction needs to set different optimizing parameters for different Raman spectra, which will be inconvenient to users. In this paper, a locally.dynamically moving average algorithm (LDMA) for the fully automated baseline correction is presented and its basic ideas.and steps are demonstrated in detail. In the LDMA algorithm the modified moving averaging algorithm (MMA) is used to strip the Raman peaks. By automatically finding the baseline subintervals of the raw Raman spectrum to divide the total spectrum range into multi Raman peak subintervals, the LDMA algorithm succeed in dynamically changing the window half width of the MA algorithm and controlling the numbers of the smoothing iterations in each Raman peak subinterval. Hence, the phenomena of overcorrection and under-correction are avoided to the most degree. The LDMA algorithm has achieved great effect not only to the synthetic Raman spectra with the convex, exponential, or sigmoidal baseline but also to the real Raman spectra.