The quantitative structure-activity/toxicity relationships (QSAR/QSTR) of 112 fluorine-containing pesticides were studied by means of the DISCO and CoMFA method. Using the 78 compounds among them chosen randomly as a training set to build up the QSAR/QSTR model, it is found that the contribution to activity from steric effect is 60.4%, and that from electrostatic effect is 39.6%. As to the QSTR model, the contribution from steric effect is 59.2%, and that from electrostatic effect is 40.8%. The crossvalidated R^2 are 0.652 and 0.611, and the non-crossvalidated R^2 are 0.982 and 0.977, and the values of variance ratio F ar 463.6 and 362.9 for the activity and the toxicity respectively. It was proved that the two models make good prediction for activity and toxicity. These models are also used to predict the activities and toxicities of other 34 compounds in the test set. The values of ratio of standard error and extremal length obtained, s/△γ, are 10.4% and 6.4% for activity and toxicity respectively. Finally these models are improved by adding more compounds into the training set. A new molecule with higher activity and lower toxicity is designed using the last model.

Key words: PESTICIDES, FLUORINE COMPOUNDS, QUANTITATIVE STRUCTURE ACTIVITY RELATIONSHIP, THREE DIMENSIONAL STRUCTURE