Research on individual differences in speed of learning has suggested that forgetting rates could be different for fast and slow learners. Studies have shown either no difference or slower forgetting over time for fast learners. The present study extends this area of research by investigating the possibility that fast and slow learning are differentially vulnerable to interference. Based on neural network models and the encoding variability hypothesis, two novel hypotheses were built and tested in two experiments by a paired-associates task. The hypotheses suggested that fast learning will be more prone to interference when similarity of the learning material is high. Hence, an interaction of learning speed and interference (i.e., similarity) was predicted. Experiment 1 (N = 22) compared retention of Chinese characters for fast and slow learning (both subject and item-specific speed) by manipulating similarity (high vs. low) of the characters learned. Results of Experiment 1 were inconclusive. Experiment 2 (N = 21) had the same basic design as Experiment 1, but included a number of procedural improvements. Interactions in the predicted direction were found both when comparing learning speed between subjects as well as for item-specific speed. However, only the interaction of between-subjects learning speed and similarity was significant. A joint analysis, including data from both experiments, yielded significant interactions for both subject speed and item-specific speed, indicating that the lack of a significant interaction of item-specific speed and similarity in Experiment 2 was probably due to the low sample size. The findings are discussed in relation to previous research on individual differences in learning speed and forgetting.