This study examines the pronunciation errors of onset stop consonants made by English learners when reading sentence-level scripts in Korean. The analysis is grounded in a dataset consisting of 7,650 sentences, encompassing a total of 319,777 syllables. This dataset was extracted from a corpus named “A Collection of Foreigners' Korean Speech Data for AI Training.” English speakers, accustomed to a two-way voicing contrast in their L1 stops, often struggle with the three-way contrast in Korean stop sounds. We conducted a data analysis to explore two main aspects: the level of pronunciation difficulty across the three categories of Korean stop sounds as reflected in error rates and prevalent substitutes for mispronounced stops. Notably, tense stops exhibit the highest error rates, contrary to earlier studies emphasizing lax stops as the most challenging for English learners of Korean (Jo 2005, Kim et al. 2006, among others). The error rates in the pronunciation of Korean onset stop sounds exhibit an inverse relationship with the degree of aspiration in onset stops, categorized as unaspirated (tense), slightly aspirated (lax), and heavily aspirated (aspirated). This inverse correlation is also reflected in error realization patterns, where aspirated stops, which showed the lowest error rates, frequently substitute mispronounced sounds. The findings imply that the production error is associated with the amount of aspiration in the target sounds that claimed to be closely related to the length of VOT (Awoonor-Aziaku 2021, Kim 2021, Liu et al. 2007). Additionally, there is a tendency for word-initial lax stops to differentiate from word-internal lax stops, indicating English speakers’ sensitivity to voicing. Our corpus analysis findings align with previous studies by Kong et al. (2022), among others, highlighting the dominant role of VOT as a cue in English learners of Korean learning Korean stops.