This study aims to investigate the characteristics of Korean high school students at four different proficiency levels, by identifying describing verb errors in their spoken English. Apart from frequency-based error analyses, this study scrutinized the error context where a particular rule should or should not apply. 400 written performance of Korean high school learners extracted from a large scale English diagnostic speaking test were compiled into 28,000 words of the English Spoken Corpus of Korean High School Learners (ESCKHSL). The corpus was error-tagged manually by two native speakers of English. The analyses of the corpus suggest the following findings: (i) some error types, such as Verb Omission (VUO), Verb Addition (VUA), Verb Choice (VC), Subject Verb Agreement (VA), and the ‘-ing’ form are frequently found at all the levels; (ii) the use of the ‘-ing’ form can be useful to discriminate between learners at different levels; and (iii) the majority of the learners overgeneralize the SV Agreement rule regardless of their levels, but higher level learners mostly over-apply the rule with the verb ‘to be’. The findings show that error contexts can provide useful information on the characteristics of Korean learners at different levels. A further study on the written performance of the same learners would be useful to understand the language ability of the Korean high school students at different proficiency levels.