This paper investigates the semantics of three Japanese evidential markers — rashii, yooda and sooda. These three words are often used in similar situations and interpreted in English as 'it seems', 'it appears', or 'it looks like'. The expressions are semantically closely related, but sometimes they are not interchangeable. Thus the question arises how to articulate the subtle differences between them. Previous studies have attempted to explicate the differences by using explanatory terms such as 'direct' or 'indirect' to describe the content of information, and 'objective' or 'subjective' to describe the attitude towards the information. While these terms are convenient to capture the meaning simplistically, they illustrate only part of the words' usage, and also the definitions apply equally well to other evidential markers. This study is the first explication of the meanings of these markers using metalanguage and the framework of the Natural Semantic Metalanguage Theory (NSM Theory) proposed and developed by Anna Wierzbicka and colleagues (Goddard & Wierzbicka 1994, 2002; Peeters ed. 2006; Goddard ed. 2008). By analyzing the deficiencies of the previously presented definitions, and examining actual usage examples drawn from modern Japanese literature, the article applies semantic primes to explicate the meanings of rashii, yooda and sooda. The meanings of each expression are illustrated by cognitive scenarios such as 'I think I can say something like this about X', or 'I think this about X at the moment'. The resulting semantic formulae clarify the differences between the three expressions. They also have utility for assisting second language learners in decisions about using the three terms.