Working memory (WM) is a cognitive function that is engaged in several everyday tasks. WM performance predicts performance in diverse other cognitive functions. Additionally, WM decline at old age is associated with age-related impairments in others cognitive functions, thus affecting autonomous performance of everyday tasks. It has been shown that WM can be improved with training interventions, and evidence has accumulated showing that also other cognitive functions can profit from WM training. The transfer findings indicate that WM training might enclose a mechanism to improve cognitive functions in general. Even though there exists a growing body of evidence on the possibilities to improve cognitive functions with WM training in different populations, the exact mechanisms of training and transfer have remained unclear. In the current dissertation I examine the prospects and precise mechanisms of WM training with four studies using the bi-modal dual n-back paradigm. I showed that dual n-back training improved performance in various tests tapping executive functions. I could also demonstrate that the mechanisms underlying transfer result from an improvement in a specific process tapped by the training task rather than in the boosting of a general cognitive ability. Consequently, transfer can occur to tasks if they engage the same specific process. Additionally, I provided primary evidence that only 16 sessions of WM training produces microstructural changes in white matter pathways connecting brain regions that support WM functions. I also showed for the first time that age-related differences in WM performance between young and older adults can be compensated for after only 16 training sessions. The findings of the present dissertation are discussed in relation to the flexibility of cognitive functions and the plasticity of the underlying neuronal substrate; additionally, new conceptions to models of training and transfer mechanisms are presented.