Vocabulary development is an ongoing process that begins in early infancy and continues into early adulthood. Moreover, early word-learning abilities are strongly related to later literacy performance. Fast mapping (FM) is a critical word-learning mechanism for early vocabulary growth in 2–3-year-old toddlers. Aside from FM, retention is another crucial ability that enables children to retain and integrate a novel word into their lexicon. In the past, word-learning studies focused on how children learn a noun and were mostly based on English acquisition. According to the literature, word class can influence how children learn a new word; for example, among toddlers, learning the word for a novel action is more difficult than learning the word for an object, implying that learning verbs is harder than learning nouns. Some research has explained that many concrete nouns refer to naturally discrete referents. By contrast, even verbs with fairly concrete meanings (such as motion verbs) are more ambiguous than nouns. Owing to the high-frequency input of verbs and their salient position in a sentence, Mandarin is considered a verb-prominent language. Researchers have also discovered that Mandarin-speaking toddlers have a higher ratio of verbs in their lexicon compared with similarly aged speakers of other languages. In the current study, we investigated (1) the scope and composition of toddlers’ lexicons; (2) the word-learning performance (FM, Retention, and Production) of 25–45-month-old Mandarin-speaking toddlers; (3) how word class affects toddlers’ performance in learning new words; and (4) the relationships between toddlers’ intrinsic word-learning abilities and their extrinsic lexicon. This study recruited 40 developmentally typical Taiwanese toddlers who met the following criteria: (1) first language was Mandarin and (2) were not developmentally delayed or physically or mentally impaired. The average participant age was 37 months (standard deviation [SD] = 5.15). We used a standardized assessment, namely the Mandarin Chinese Communicative Development Inventory, Taiwan version, to assess the word composition and lexicon of individual Mandarin-speaking toddlers participating in the present study. We used a self-designed word-learning task adapted from referent selection design (Horst et al., 2010; Lu & Tsao, 2014), which included three phases, FM, retention, and production. Two versions of the task were designed—noun and verb— to investigate whether nouns are easier to learn than verbs when both word types are presented in comparable sentence contexts, controlling for the number of exposures. Each version of the task included four familiar and four novel words to examine how novelty affects word-learning. The procedure began with the noun condition and after a 10-min break, proceeded to the verb condition. Each condition contained a learning event, FM, production, and retention phase. During the learning event phase, a laptop was employed to present toddlers with the referent picture. Three prerecorded instruction sentences were played; for instance, “zhe shi (this is) ‘referent object’,” “you yi ge (there is one) ‘referent object’,” and “‘referent object’ zai zhe li (is here)!” These three sentences are all examples of common phrases used in daily conversational Mandarin. After three exposures to the word, the study transitioned to the FM phase in which three pictures (including a correct referent object’s picture) were presented on the laptop and the examiner asked the toddler, “In the three pictures, which one is the ‘referent object’?” In the production phase, the referent object´s picture was presented, and the examiner asked, “What is this?” to encourage the toddler to say the novel word without prompting. After a 5-min break, without any recap or practice of the word, the Retention phase began. In this phase, three pictures were presented to the toddler, and the examiner asked, “In the three pictures, which one is the ‘referent object’?” The same procedure was applied when verbs were used, but a 20-s prerecorded referent action film was presented in the learning event phase, and referent action GIFs were presented in the FM, production, and retention phases. The number and sequences of the presented stimulus in the aforementioned phases were well-controlled. Descriptive statistics, including percentages and means, one-way repeated-measure ANOVA, and statistical significance tests were used to analyze the data. In addition, Pearson’s correlation and linear regression were respectively used to examine the relationship and these variables’ predictability of each other. We determined that the average lexicon size for Mandarin-speaking toddlers was 536.67 words [SD = 128.05], and on average, nouns accounted for 50% [SD = .24], verbs and adjectives accounted for 25% [SD = .02], and close words accounted for 9% [SD = .02] of their lexicon. Our results are consistent with the literature showing that Mandarin-speaking toddlers exhibit “noun bias” in their early lexicon (Chen & Liu, 2014). Second, in the word-learning task, we found that Mandarin-speaking toddlers at this stage have well-developed noun and verb FM abilities. Their accuracies for novel word-learning, both nouns and verbs, exceeded 80%. However, they demonstrated little ability in the retention (noun acc = .64, SD = .25; verb acc = .44, SD = .3) and production phases (noun acc = .09, SD = .12; verb acc = .08, SD = .21). These findings indicate that 2–4-year-old Mandarin-speaking toddlers can successfully map a novel word onto a referent object or action but fail at retaining novel words after a 5-min retention phase. Additionally, we determined that the learning performance for verbs in the retention phrase was significantly worse than that for nouns [F(1, 38) = 16.97, p < .001, η2= .31]. These results indicate that FM and retention can be seen as two different mechanisms in word-learning and also suggest that learning a verb or action is more difficult than learning a noun or object, which is consistent with the literature. Finally, there was a medium positive correlation between verb learning and toddlers’ lexicon in both the FM (r = .482– .564, ps < .01) and retention (r = .336– .485, ps < .05) phases, which was not seen during the noun learning task. Moreover, individuals’ verb learning ability was a significant predictor of lexicon size (r2 = .28, p < .001), and lexicon size could also predict verb learning ability (r2 = .28, p < .001). This suggests that toddlers’ verb learning ability and their lexicon have bidirectional predictive power, meaning that toddlers who have larger lexicons demonstrate better verb learning ability and those who have better verb learning ability also tend to learn more new words. Under daily word-learning conditions, verbs usually appear with other words (e.g., known words or pronouns) in a sentence. For instance, when a novel verb appears in a structural argument with other types of words, “syntactic bootstrapping” can facilitate verb learning. In summary, our findings indicate that 2–4-year-old Mandarin-speaking toddlers have well-developed FM ability for both nouns and verbs but only a small proportion can retain the newly-learned words. Retaining and integrating a verb’s (action) representation into the lexicon can be more challenging than doing so with a noun (object). We conclude that in the early vocabulary learning stage, toddlers who can utilize verbs or have a large verb vocabulary at their command are privileged in word-learning and that verb learning ability plays a crucial role in early vocabulary development for Mandarin-speaking toddlers. It would be beneficial to integrate the assessment of novel word-learning processes of individuals with typical and atypical development into the evaluation of early language development and disorders.