In an embodiment of the invention, a method for coding a moving picture comprises: dividing an inputted picture signal into fractionized signals respectively corresponding a plurality of pixel blocks; generating a prediction signal and coding mode information from each of said fractionized signals, for each coding mode; generating a prediction residual signal, from the prediction signal and said each of fractionized signals; estimating a first-estimate coding cost that is for coding the prediction residual signals for said each coding mode, from the prediction residual signals and the coding mode information; determining a candidates' number in accordance with step width of quantizing; selecting the coding modes having smallest ones of the first-estimate coding costs as candidates; estimating a second-estimate coding cost by coding the inputted signal and thereby finding a coding distortion and a code amount; and employing one coding mode from the candidates on basis of the second-estimate coding cost.