Spatial distributions of chemiluminescence from OH, CH and C2 radicals in flames have been drawing great attention in the literature since the ratios of the local intensities OH/CH/C2 could be used to monitor such characteristics as fuel air ratio, completeness of combustion, rate of combustion, blowout. With the rapidly growing capability of optical sensor technologies there is an increased interest to develop an accurate tomographic reconstruction technique for combustion control. A specific difficulty of flame tomography is the character of the source function to be reconstructed: the function describing the spatial distribution of chemiluminescence has narrow high-gradient peaks. The goal of the present study is to propose a reliable reconstruction technique applicable to axisymmetric and non-axisymmetric flames. Computing algorithm based on the maximum entropy (MENT) concept in combination with data preprocessing procedure has been developed. This algorithm is optimal in conditions with limited observation directions. A new method based on local regularization has been proposed for data processing to achieve more accurate reconstruction of the peak intensities and thicknesses of chemiluminescent zones in flames. The developed approach is tested in reconstruction of OH, CH and C2 radical chemiluminescences in the axisymmetric propane air flame of a Bunsen-type burner. (C) 2012 The Combustion Institute. Published by Elsevier Inc. All rights reserved.