Laser absorption spectroscopy (LAS) tomography is well-proved in combustion diagnosis but has difficulty especially in the simultaneous imaging of multi-species concentrations. A multiple species imaging method from single species LAS tomography was proposed on the basis of computational fluid dynamics (CFD) and transfer learning. CFD simulation of the methane/air flat flame was conducted to reveal the relationship among multiple species. A back propagation (BP) neural network was pre-trained with the dataset obtained from CFD simulation to predict projection values of OH mole fractions from H2O absorption lines at 7185.6 cm-1 and 7444.4 cm-1. The measurement of flat flame by a single wavelength planar laser-induced fluorescence (PLIF) fused LAS tomography system was conducted for network fine-tuning and experiment verification. Distributions of OH mole fractions in lean-burn conditions and nearly complete combustion conditions were quantitatively reconstructed well, while annulus profiles in fuel-rich conditions were qualitatively retrieved. Reconstructed images with two-fifth experiment data used in the network fine-tuning showed a 31.3% decline in image error compared to those without fine-tuning. This proposed method enables LAS tomography of multiple species via only one species with enough measured projections, and also shows potential in image error reduction by introducing more projections.