The method includes steps: (1) placing fruit sample inside sealed container; after top air portion reaches balance, sampled top air portion is inducted to sensor array reaction chamber; response signal is obtained when reaction takes place between sensor array and top air portion; response signal of sensor is ratio between resistance R after sensor contacts top air and resistance R0 when sensor passes through clean air, i.e. S=R/R0; (2) carrying out detections including compactness, sugar degree, and acidity for detected fruit; (3) using multiple linear regression, major constituent regression, least squares regression, and artificial neural network to build mathematical model of relation between the response signal of sensor and compactness, sugar degree, acidity of fruit sample. The invention extends detection range, lowers interference, increase sensitivity, reliability, and repeatability.