Automatic crystallization systems allows to make about 100–1000 and even more crystallization experiments per hour. However only part of the experiments result in crystals and only some of the crystals are of the special interest. Usually such cases are detected manually by a person who recognizes crystals in drops by their appearance. In order to improve the effectiveness of crystallization experiments, the automatic detection systems, based on image analysis approach, have been invented recently, but the recognition quality of these systems still needs significant improvements.

In the present work we have discovered the problem of recognition of crystal drops in general and some special cases, which are of a special interest, in particular. Several approaches, based on such methods as eugenfaces, wavelet transformation and texture analysis, have been used to get relevant image features. Principal component analysis was employed for exploratory data analysis, soft independent modeling of class analogue (SIMCA) and support vector machines (SVM) have been utilized for pattern recognition. The detailed comparison of the results is shown.