Visual working memory (VWM) is limited in the amount of visual information it can retain and the rate at which it encodes that information. VWM capacity is robustly correlated with cognitive measures, which is one reason that researchers are interested in factors that enhance memory performance and whether training is one such factor. Observers performed a change-detection task with both familiar English letters and the unfamiliar Brussels Artificial Character Set (BACS)—an artificial alphabet matching the number of junctions, strokes, and terminations of English letters. We used the Courier New font similar to BACS in perimetric complexity—an objective estimate of the number of visual features contained within the letter. The delay between memory array and mask was varied, allowing measurement of VWM capacity as a function of encoding time. We found a higher encoding rate and capacity for English letters relative to the BACS, reflecting an effect of familiarity. We then used a training protocol shown by Blalock (2015) to produce enhanced change-detection performance for random polygons. During training, observers were presented a target BACS letter for a short duration before having to recognise the target in an array of four letters. We found high recognition accuracy following the training, showing observers became familiar with the trained BACS letters. However, memory performance on a change-detection task did not differ between trained and novel BACS letters, and was significantly worse than the highly familiar English letters. This suggests a boost to memory performance requires more substantial experience with stimuli.