The nature of capacity limits within human visual working memory (VWM) remains the subject of controversy: while the capacity-as-objects account predicts that what loads VWM capacity is solely the number of objects maintained, irrespectively of the number of visual features that need to be stored for each object, the capacity-as-features account predicts that also (or – primarily) the total number of features maintained in VWM loads its capacity (and leads to decreased performance). We present novel simulations of a VWM task, using our existing, oscillatory computational model that describes the binding of features into objects as resulting from the proper synchronization and desynchronization of rhythmical changes in neuronal activity. The model predicted (in line with wide evidence) that VWM performance decreases with the increasing number of objects, but also decreases (although not as sharply as predicted by the capacity-as-features account) as a function of increasing number of features. The model attempts to explain what precise characteristics of oscillatory dynamics stand behind such two sources of VWM limitation. However, the complete pattern of the model’s predictions remains yet to be examined empirically.