How much, how high, how fast, how many? We routinely evaluate information based on abstractions of magnitude, be it in our everyday judgments or in psychophysical experiments. In this talk, I will address in particular two questions: (i) how is magnitude information kept in working memory (WM), and (ii) how do humans integrate abstract magnitude in comparative decisions? I will review evidence that keeping stimulus information ‘active’ in WM can involve high-level representations of magnitude in frontal areas, even in putatively simple psychophysical tasks. These findings will be discussed in light of contemporary models of WM function and therein, I will outline a potential role of beta-oscillations in endogenous content updating. In the second part of my talk, I will present new evidence for human biases in integrating numerical magnitude in decision making. One study showed that during comparison of approximate number, the weight of sequential inputs is not constant, but fluctuates at the rhythm of endogenous low-frequency signals over parietal cortex. Lastly, I will illustrate a novel, systematic bias in numerical comparisons, which, paradoxically, may act to maximise accuracy in human observers.