The parameter inverse algorithm of rainbow measurement was studied. The mathematic model was built and a novel inverse algorithm was presented. The new algorithm removes noise using the technique based on empirical mode decomposition（EMD）which can filter out the noise interference under low signal-to-noise ratio(SNR) condition and effectively avoid signal loss and shift. A feature point extraction technique was presented, to compress the inversion calculation. An optimal point search algorithm based on Debye theory was proposed which can rapidly find the relatively exact inversion parameter, and more accurate parameter was obtained using Mie theory. The new algorithm greatly improves the accuracy of parameter inversion, especially reduces the refractive index inversion error and expands the application scope. The algorithm was validated by numerical test in different noisy environment for water droplets with diameter range from 35 μm to 900 μm and refractive index from 1.32 to 1.34. It is showed that the inversion diameter accuracy can reach 1 μm, and refractive index can reach the third place after decimal point with the SNR above 40 dB. When the SNR drops to 5 dB, the maximum inversion error of diameter is less than 10％ and error of refractive index is less than 0.1％.