The human brain has a considerable division of labour. Every mental state is encoded in a complex activity pattern across the wide network of brain regions. For this reason, it is necessary, to obtain full brain coverage in order to decode a wide range of mental states from brain activity. Currently, the only possibility to obtain full coverage of the human brain, including subcortical and deep cortical regions is functional magnetic resonance imaging (fMRI), albeit at the expense of temporal resolution. This talk will provide an overview of mental state decoding using fMRI spanning such diverse functions as visual perception, emotions, action plans and memories. The easiest case is visual perception because a large proportion of the cortical surface is dedicated to vision and thus considerable information can be recovered. Related sensory contents, such as those in visual imagery and in visual working memory can also be decoded from visual cortex, in line with the idea that the brain re-uses the most dedicated and specialized regions when encoding cognitive contents. Decoding of emotional states is also possible and reveals a distributed set of regions rather than a few, high specialized areas (such as the amygdala for fear). Action plans are also accessible to fMRI decoding, even though the temporal resolution is not sufficient for realistic real-time control of brain-computer interfaces. The talk will also present a number of key challenges in our field such as individual differences in cortical representations, the vast number of potential cognitive states or the validity of applied decoding approaches (e.g. for neuromarketing or lie detection). Thus, mental state decoding from fMRI signals is very powerful within its natural limitations, but most real-world applications are still under development.