Contamination of drinking water with arsenic is a recurring problem in both industrialized and developing countries. Supplies for large populations can have concentrations much higher than the permissible levels, set by the World Health Organization (WHO) to 10 μg As/L for most European countries and the United States and to 50 μg As/L elsewhere. As arsenic analysis requires high-end instruments, which are largely unavailable in developing countries, bioassays based on genetically engineered bacteria have been proposed as suitable alternatives. Yet, such tests would profit from better standardization and direct incorporation into sensing devices. The final objective of this work was to develop a microfluidic device in which bacterial bioreporters could be embedded, actively maintained for at least one week, exposed to arsenic and which allowed direct detection of the reporter signal produced, as a further step towards a complete miniaturized bacterial biosensor. The signal element in the biosensor is a nonpathogenic laboratory strain of Escherichia coli, which produces a variant of the green fluorescent protein after contact to arsenite and arsenate. To reach the stated objective, we proposed two different solutions. The first one consists in the encapsulation of E. coli bioreporter cells in agarose beads and in their incorporation into a microfluidic device, where they are captured in 500 x 500 μm2 cage and exposed to aqueous samples containing arsenic. Cells in beads frozen at -20◦C in the microfluidic chip retained inducibility for up to a month and arsenic samples with 10 or 50 μg As/L could be reproducibly discriminated from the blank in less than 200 minutes. In the second approach, we directly captured free cells against a filter wall and for their active maintenance we integrated a microchemostat on chip. Because of a lack of robustness of the microfluidic device, we could not maintain cells on chip for more than one week, but we showed that it is possible to have a biosensor in which the sensitive element is continuously renewed and, when needed, exposed to the target chemical. In the last part of this thesis, we studied the relation between arsenite transport and reporter signal production. We observed the formation of extensive gradients of reporter signal intensity as a function of distance to the inflowing sample, arsenite concentration and flow rate, and we attempted to explain their nature by a modeling approach. We think that the devices proposed in this work constitute a crucial step forward in the direction of an inexpensive and robust in-field arsenic detection system.