Time-series vegetation index data are contaminated with residual noise and cannot be used in land-cover change detection and crop yield estimation directly. To remove noise effectively, researchers have developed a series of methods for vegetation-index data reconstruction. Function curve-fitting methods are popular in the reconstruction of time-series vegetation index data and have been widely applied in many fields. Different function curve-fitting methods have specific adaptabilities to different geographical environments. In practice, researchers usually have to compare many function curve-fitting methods and select the most suitable one according to the characteristics of regional time-series vegetation index curve fluctuation. Therefore, the means of comparing different function curve fitting methods objectively and quantitatively is very important. Based on ten-year MODIS EVI data of evenly distributed sample areas and its quality control data from the Qinling Mountains, the evaluation method for EVI time-series data reconstruction was discussed and developed in this study. The new evaluation method can compare different function curve fitting methods objectively and quantitatively on two important aspects. One is the function curve-fitting effect under the disturbance of noise points, and another is the ability of retaining original high-quality data. In this study, we used EVI time series data of the sampling area in the Qinling Mountains to analyze the stability of the maximums and minimums of the EVI curves and found that the maximums are more stable in the EVI time series data than the minimums. Then we modified the form of the single logistic model on the basis of the above analysis. Finally, the Maximum optimization Logistic function fitting method (MSL) was proposed to improve the accuracy of the EVI time series reconstruction with large noise in complex mountains. In this study, a new evaluation method was used to compare the Asymmetry Gauss function fitting method (AG), Double Logistic function fitting method (DL), and Single Logistic function fitting method (SL) with the Maximum optimization Logistic function fitting method (MSL). The results show the following: (1) For the function curve fitting effect under the disturbance of the noise points while maintaining original high-quality data, AG showed better results in the treatment of several pixels. (2) Compared to the AG and DL, the fitting effect of SL and MSL is more significant. They not only undisturbed by the noise points but also have a stronger ability to maintain the original high-quality data than AG and DL. (3) Compared to the other methods, MSL was found to be more applicable for EVI time-series data reconstruction in complex mountains with large noise.