The automated detection and measurement of oil drops in multiphase fermentation systems are important for mass transfer analysis. A novel computer technique for automated detection of oil drops in images is presented in the context of a stirred tank containing a three-phase water-oil-air dispersion. The technique is an original feature extraction transform designed for the detection of objects with a characteristic appearance. The proposed transform, denominated gradient-direction-pattern (GDP) transform, utilizes naturally occurring patterns in the orientation of the local gradient appearing in test images. The GDP transform was used to demonstrate the feasibility of automatically estimating oil drop-size unbiased distributions which is an important task in the chemical and other related industries.