This paper studies domain-dependent implicit sentiment expressions in earthquake-related social media texts. In sentiment analysis, domain-dependent implicit sentiment expressions are difficult to estimate and recognize, which often results in the depreciation of system performance. In this study, we analyzed two types of implicit sentiment expressions that may occur in the causal clause of the sentence conveying ‘fear’ sentiment: {simple predicate} type such as ‘collapse’ and {complex predicate} such as ‘blow up’. When these clauses appear without main clauses that express explicit sentiments such as ‘anxiety’ or ‘panic’, they imply the omitted sentiment and thus implicitly express certain sentiments by themselves. The DECO T-Crawler platform was used to collect earthquake-related data in social media texts and the DECO Korean dictionary and local-grammar graph formalism were used to the corpus. The performance of resources built in this study showed 84.5% accuracy in a test corpus, which is about 23% higher than the accuracy expected when using only sentiment-conveying words.