A genetic-algorithm-based selective principal component analysis (GA-SPCA) method is proposed and tested using hyperspectral remote sensing data and ground reference data collected within an agricultural field. The proposed method uses a global optimizer, the genetic algorithms, to select a subset of the original image bands, which first reduces the data dimension. A principal component transformation is subsequently applied to the selected bands. By extracting features from the resulting eigenimage, the remote sensing data, originally high in dimension, will be further reduced to a feature space with one to several principal component bands. Subsequent image processing on the reduced feature space can thus be performed with improved accuracy. Experiments were conducted using three sets of ground reference data: corn chlorophyll content, corn plant population, and various corn hybrids. The results showed that with GA-SPCA, the number of original bands used for principal component analysis (PCA) could be reduced to 17, 26, and 25 from a 60-band hyperspectral image, respectively. In all cases, the correlation coefficients between image and ground reference data were greater when using GA-SPCA than that for PCA results with all original bands. This indicates that bands with no contribution to a specific application were removed prior to PCA. The variance related to a specific application within the image was transformed with more emphasis by using bands sensitive to that application. The selected bands can also provide useful information for future imaging sensor development.