Background: Seed size and number are important plant traits from an ecological and horticultural/agronomic perspective. However, in small-seeded species such as Arabidopsis thaliana, research on seed size and number is limited by the absence of suitable high throughput phenotyping methods. Results: We report on the development of a high throughput method for counting seeds and measuring individual seed sizes. The method uses a large-particle flow cytometer to count individual seeds and sort them according to size, allowing an average of 12,000 seeds/hour to be processed. To achieve this high throughput, post harvested seeds are first separated from remaining plant material (dust and chaff) using a rapid sedimentation-based method. Then, classification algorithms are used to refine the separation process in silico. Accurate identification of all seeds in the samples was achieved, with relative errors below 2%. Conclusion: The tests performed reveal that there is no single classification algorithm that performs best for all samples, so the recommended strategy is to train and use multiple algorithms and use the median predictions of seed size and number across all algorithms. To facilitate the use of this method, an R package (SeedSorter) that implements the methodology has been developed and made freely available. The method was validated with seed samples from several natural accessions of Arabidopsis thaliana, but our analysis pipeline is applicable to any species with seed sizes smaller than 1.5 mm.