The experiment was conducted in 1992 to find out the approach to the development of prediction model of weed population in paddy fields. The weed seeds of 88% over were separated from the soil by using 50% solution with specific gravity 1.34. The weed seeds which were floated on the solution due to the difference of specific gravity between soil particles and the seeds were effectively withdrawn by using a vaccum pump attached with an aspirator. The seeds withdrawn together with solution were taken by filtering with a nylon net of mesh. The pressing method was more efficient and practical for the viability test of weed seeds separated from the soil compared with the germination test and the TTC test. For the prediction of weed population by the number of weed seedlings emerged at the sampled soil, the sampling method of 0-10cm deep at 5-6 sites per field was applicable. At the prediction method by the number of seedlings emerged, the smaller the seed sizes, the lower the prediction coefficients of weed species. It was considered that the prediction method by the number of seedlings emerged was more practical than the prediction method by the number of seeds separated from sampled soil, in relation to similarities to weed population, time and expenses required for examining, technical difficulties and applicability of weed species.