Using spectral analysis techniques, this research aims to estimate kalium content of apple flowers. Firstly, the original spectral reflectance (Ri) of apple fresh flowers at full-bloom stage of Red Fuji and flower kalium content (KC) were measured, and correlations between kalium content and Ri and eight kinds of spectral transformation data (Ri/、Ri//、1/Ri、(1/Ri))/、lg(1/Ri)、(lg(1/Ri))/、lgRi、(lgRi)/) were analyzed. Sectral reflectance variant with the highest correlation coefficient was discovered. Secondly, stepwise regression analysis of kalium content and the corresponding spectral reflectance variant with higher correlation coefficients was carried out to determine sensitive characteristic wavelength. Estimation models for kalium content of apple flowers were established with sensitive wavelength. The best estimation model was determined by model optimization and validation. The results showed that the highest correlation coefficient appeared between kalium content of apple flower and the first derivative of spectral reflectance Ri/ and the second highest relevance was for 1/Ri, lg (1/Ri) and lgRi. Their absolute correlation coefficient peaks (valleys) appeared at wavelength of 669 nm, 952 nm, 1164 nm, 1442 nm, wavelength of 351 nm, 352 nm, 362 nm, 366 nm, wavelength of 351 nm, 366 nm, and wavelength of 351 nm, 366 nm. The sensitive wavelength were 669 nm, 1442 nm, 352nm, 351nm, 351nm, which were defined by stepwise regression analysis. The established regression models had better linear trend, and the estimation model with the first derivative of spectral reflectance variable had the maximal correlation coefficient up to 0.6113. The estimation model was tested by the test data samples, with its fitting equation coefficients of determination (R2) of 0.6955, root mean square error (RMSE) of 2.7, and the relative error (RE) of 4.9%. It was indicated that the estimation model of kalium content of apple flower with the first derivative of spectral reflectance had the highest accuracy, which was the best estimation model with a certain degree of stability and applicability. Our results provided theoretical basis and reference value for the rapid estimation of apple blossom nutrient content and real-time nutritional diagnosis.