Early research on the representation and processing of information in bilingual memory debated whether the two languages were stored and accessed together or separately in memory. In recent years, it is an important issue for how the connection between words and their meanings is developed with increasing proficient in the L2. Results from previous studies led to the conclusion that the representations models were different between high proficient bilinguals and less proficient bilinguals. However, the conclusion has been supported by some experiments based on explicit memory tasks. Therefore, we conducted the present study to explore the lexical and conceptual representation model of the less proficient bilinguals, using cross-language repetition priming which is based on implicit memory tasks. In the present study, Participants included 165 junior students with English as their second language. The students were non-English-majors whose scores of CET band 4 were between 280 and 350. All experiments and data collection were managed by E-prime on computers, to assure accurate reaction time measurement. Participants were instructed to complete the lexical decision task in experiment 1, but the conceptual decision task in experiment 2.There were 2 phases in each experiment, the study phase and the test phase. The reaction time and accurate rates were analyzed by SPSS 17.0, using a two-way repeated measure ANOVA with language group as a between-participant factor (same language, cross language) and trials type as a within-participant factor (studied trials, non-studied trials). There was no cross language repetition priming effect in experiment 1 in cross language condition. The response time to the words studied was significantly faster than that to the words unstudied in experiment 2 both in same language condition and cross language condition. In experiment 2 there was symmetrical in within-language repetition priming effect, but the asymmetrical pattern of between-language. In other words, the response time to the words studied was significantly faster than that to the words unstudied in English-Mongolian condition, but no cross language repetition priming effect in Mongolian-English condition. The result of the two experiments indicated that, for less proficient bilinguals, lexical representations are separate. To our surprise, conceptual representations for translation equivalents are shared but asymmetrical. Further, our results support the Revised Hierarchical Model.