ABSTRACT This paper introduces a new Japanese courseware program, Banzai Readings, and presents an empirical study based on the program. Banzai Readings provides reading texts accompanied by glosses to enhance second-language learners' reading comprehension. Several studies support that marginal glosses are more effective than no gloss in learning second-language vocabulary (e.g., Hulstijn et al. 1996, Watanabe 1997). Since marginal glosses can be presented in different formats, it is of interest to determine which formats are more effective than others. This study investigates the effectiveness of two types of glosses using Banzai Readings. The single-gloss version of the program provides a single English translation for each target vocabulary item or target grammatical structure occurring in the reading text. The multiple-choice version of the program provides two possible translations in a multiple-choice format, followed by immediate feedback in response to learners' selections. The result of the study is that the multiple-choice format is significantly more effective than the single-gloss format. The study illustrates how educational software design decisions can be usefully guided by empirical studies of the relative effectiveness of different versions.