The research described the proficiency degree, selection of krama lexicon, and deviation of Javanese speech levels usage. The data is taken from krama lugu dan krama inggil that is gathered through structured and in-depth interviews. Sociolinguistics approach with the the utilization of both quantitative and qualitative method is used in analysis data. Quantitative analysis is used to measure krama lexicon’s proficiency degree. Meanwhile, qualitative analysis is utilized to reveal some factors that become the background of krama lexicon selection and deviation of speech levels usage in a polite utterance. The qualitative analyze employs method: bagi unsur langsung (BUL), permutation, and substitution. The result shows that the proficiency degree of krama lexicon of Semarang coastal society Javanese speaker’s is 38,1% (qualified as: less good). Another findings of the research is the lexicon krama’s deviation in Semarang coastal’s Javanese language utterance. The deviation is measured with the standard parameter of Javanese Language (prescriptive), such as: (1) the overlaps of Javanese language utterance, (2) self-kramanisasi, and (3) interference of Javanese language and Indonesian language. That deviations caused of an error in language inheritance, a low proficiency in lexicon krama, and the lack of comprehension of understanding Javanese speech levels usage.