The determination of tissue optical properties is the fundamental research field in biomedical optics. The ability to separately quantify absorption and scattering coefficients of tissue based on diffuse reflectance spectrum not only helps to gain physiological and structural properties of tissue but also provide insight into the mechanisms of tissue, which leads to the improvement in non-invasive detecting, image diagnosis and photodynamic therapy. In the paper, a flexible and rapid method is developed to extract the absorption and reduced scattering coefficients of turbid medium such as human tissue with diffuse reflectance spectrum. The diffuse reflectance spectrum is measured by the system which includes a white light source, a spectrometer, and a fiber optic probe for delivery and collection of light. The collection efficiency and system transfer function are researched based on the fiber probe geometry. This paper outlines a method based on empirical forward model and non-linear modeling inverse model to extract the optical properties from diffuse reflectance spectrum. The approach includes four steps: （1） generating diffuse reflectance spectra for training inverse model; （2） training the inverse model; （3） measuring and processing the diffuse reflectance spectra of samples; （4） predicting the optical properties of samples. Since the forward and inverse models could be regarded as non-linearity, the Artificial Neural Networks (ANN) is employed to develop the forward and inverse models. The principal component analysis (PCA) is also employed in the inverse model to decompress the data dimension and suppress the spectral noise. With a single fiber optic probe and spectroscopy system, the diffuse reflectance spectrum is measured and preprocessed. The accuracy and robustness of this method are evaluated by measuring the phantoms with a wide range of optical properties. The results indicate that the absorption and scattering coefficients could be extracted accurately by measuring the diffuse reflectance spectrum of single source-detector distance. The mean RMS percentage error is 4.58% and 7.92%, respectively. As to the application of extracting concentration of different chromosphere, it is better to include the absorption peak of every chromosphere within the measuring wavelength range. This method is valid for a wide range of optical properties with the advantage of rapid measurement and simple system setup, which is important for the clinical application.