An on-site field experiment of four phosphorus fertilizer rates treatments and two cultivars of summer maize was conducted at Northwest A&F University in 2012—2014. Crop canopy normalized spectra and total phosphorus content of maize leaves were measured at jointing, huge bellbottom, silking, filling and milk stages. Nine normalized spectra wavelengths and combinations of canopy hyperspectral remote sensing of the first and second year data were chosen to establish linear, log, quadratic and exponential function regression relationships between leaf total phosphorus content and spectra parameters for each cultivar. Four models with high determination coefficients and F values of each cultivar at each growth stage were chosen to verify the models with the third year data of normalized spectra and total phosphorus content of two cultivars separately. The models with the smallest root mean square error and relative error were taken as the best models. The results showed that the spectra parameter for the best fitting regression relationship with leaf total phosphorus content were the normalized spectra of (830+880) nm, (830+940) nm and (880+1 100) nm at the jointing, bellbottom and filling stages of maize, respectively.