How the brain processes information to learn and form memories is still a widely unanswered question. To gain insights, neurons are studied under great effort in vivo. However, the applied methods often only enable monitoring subsets of neurons embedded in a complex network or lack spatiotemporal resolution for analysis on a single-cell level. In bottom-up neuroscience, small neural networks of reduced complexity are engineered by seeding primary rat neurons or human induced pluripotent stem cell derived neurons into polydimethylsiloxane microstructures on top of microelectrode arrays (MEAs) to spatially confine the location of their soma and guide neurite growth. The microstructure used in this work comprises a single seeding well per network and microchannels branching out from this node, where action potentials are recorded from the axons. Spontaneous recordings show stable response patterns for days where pre- and postsynaptic pairs can be identified. A stimulation protocol is developed, where those cells are stimulated with a positive or negative delay to induce potentiation or depression. Every stimulation phase is followed by a readout of spontaneous activity. All experiments are performed with a custom readout and incubation system, including temperature control, medium exchange and water supply to counteract evaporation. Preliminary experiments show differences in relative postsynaptic latency and spike transmission probability depending on the applied protocol. Fundamental neuroscience research is lacking a platform for stable long-term model systems with high-resolution readouts. Establishing neural networks on MEAs with reduced complexity could bridge this gap and benefit fundamental neuroscience, drug development and biohybrid technology.