分子片段是因成键环境不同而具有不同特性的原子或官能团，也可以认为是影响化合物性质的一个分子结构单元。分子片段可变连接性指数（mfVCI）是定量结构-性质相关（QSPR）研究中的一种拓扑描述符。为了检验mfVCI对同组化合物不同性质预测的适用性，本文建立了mfVCI与36个卤代苯的水溶解度（S_w）和正辛醇/水分配系数（K_ow）的线性QSPR模型，用规划求解方法得到最优解，并采用拟合优度诊断和交互验证法（内部交互验证和外部交互验证）对模型进行检验。结果表明，所建立的模型均具有较好的拟合能力（R²>0.97）和稳健性（q²>0.89），且预测能力强（q_ext²>0.95）。mfVCI除具有一般可变分子连接性指数的优势外，还能更好地区分不同成键环境中原子或官能团对同组化合物不同性质的影响，能较好地应用于卤代苯的水溶解度及分配系数的预测。

For the molecular fragments Variable Connectivity Index (mfVCI), the molecular fragments are defined as the atoms or functional groups and are also regarded as the molecular structure unit for the major influence of the property. Different molecular fragments were endued with different fragments weights. In order to test whether or not the mfVCI could be used for predicting different properties of the same molecules, the Quantitative Structure-Property Relationship (QSPR) models of aqueous solubility (S_w) and n-octanol/water partition coefficients(K_ow) for 36 halogenated benzenes based on the mfVCI were investigated. The fitting goodness of the QSPR models are mainly characterized by determination coefficient (R²), standard deviation (SD), Fischer variance ratio(F), and significant lever (P). The internal cross-validation including leave-one-out and leave-many-out methods with cross-validated correlation coefficients (q_loo² and q_lmo²) was used to evaluate the model's robustness. The external cross-validation was used to evaluate the predictive power of the models developed from the training set by correlation coefficients (q_ext²). The linear relationship was assumed between mfVCI and S_w or K_ow in here. The mfVCI was optimized by using Solver. The target function is the root mean calculated residual sum of squares for training set. The QSPR models with R² value of more than 0.97, q² value of more than 0.89, and q_ext² value of more than 0.95 have the good robustness and predictive ability as well as fitting ability. The mfVCI could distinguish the different role of same atoms or functional groups among different chemical bonding. The results show that the mfVCI could be well used in the prediction of both of aqueous solubility and n-octanol/water partition coefficients for halogenated benzenes. It could describe different properties equally well. The mfVCI maybe play an important role in the development of molecular descriptors on QSPR researches.