Aside from a report claiming that soybean seeds contain less than 1% starch, little is actually known about the genetic variation of the starch content in this important crop species. We used a starch-iodine test to identify varieties with high starch content in a soybean germplasm collection and found a total of 34 cultivars that showed a strongly positive reaction (dark color). For a more accurate quantitation of starch contents, we established a new method using a heat-resistant α-amylase and dinitrosalicylic acid (DNS) reagent. We compared the DNS method and the standard method using glucose-oxidase (GOD) with a known amount of starch as a standard. We showed that the DNS method generated results that were very highly correlated with the GOD method. In addition, we found the new method to be easier and faster to implement than the GOD method. Using the DNS method, we found several accessions in our soybean germplasm whose starch contents were 2–7%.