Processing of symbolic numerical stimuli such as Arabic numerals has been shown to activate the posterior inferior parietal lobe, especially in approximate calculation. Exact calculation also activates the dorsolateral prefrontal cortex (DLPFC) (Barbaud et al. 1999; Cantlon et al. 2006; Dehaene et at. 1999, 2004). Furthermore, lesions studies have shown that damage to the posterior inferior parietal cortex leads to severe difficulties with performing simple calculation, such as stepwise computation (Joseph 2000: 463). The same frontal-parietal network is also involved in working memory (WM), in particular with stimuli that can be recoded as structured chunks. WM demands have been shown to correlate with activation in posterior parietal cortex, DLPFC and as well as in the premotor cortex (BA 6) (Bor & Owen 2007; Bor et al. 2004; Smith & Jonides 1999). I report preliminary results from an ongoing fMRI study which involves numerical processing as well as WM and error detection. Three types of stimuli: (a) repeated chunks (x, y, z, x, y, z…), (b) smaller structured chunks requiring minimal calculation (x, x+1, y, y+1, z, z+1…), and (c) strings with an increased calculation requirement (x = x+3). The control condition consists of simple x = x+1 strings (e.g. 1, 2, 3, 4, 5, 6…). The subjects have to press a button when they detect error to the general patterns, i.e., when a number does not conform to the numerical string. Using a block design to investigate the numerical processing, all three conditions conform to the same general pattern with massive activation in the posterior parietal cortex, dorsal frontal and premotor cortex (BA 6, 8), and DLPFC, though with distinct differences in extent which reflects different demands for WM and calculation. Applying an event related analysis to examine the error detection task itself, the pattern changes. Conditions (b) and (c) both show localized and less extensive activation in posterior parietal cortex and premotor/frontal cortex. In condition (a), on the other hand, the activation covers almost the entire parietal and frontal lobes. This pattern cannot be taken to reflect WM demands. It could be related to the task since oddball detection (responding to conflict between expectation and anomalous stimuli) has been shown to activate the prefrontal cortex (Christensen et al, in prep; Huettel & McCarthy 2004). The results indicate that such activation is dependent on the nature of the oddball, that is, the type of error or anomaly.