This article deals with the automation of the process of somatic embryogenesis for the propagation of plants. An important problem is the monitoring of the embryo production process in order to decide the time to start harvesting embryos for further processing. The classification algorithm development for somatic embryos of birch (Betula pendula Roth) showed that automated recognition of embryos at different developmental stages is possible. No globular stage embryos were classified to be heart or torpedo stage and no heart or torpedo stage embryos were classified to be at globular stage. Heart and torpedo stage embryos were classified into three developmental classes by a new index that describes the relation of embryo breadth to the length of the root. The probability of classifying a nonembryo as an embryo was less than 1%, and 14% of the object classified as embryos by a human expert were discarded by the algorithm. A computer vision system suitable for automated monitoring of samples from the bioreactor was constructed.