Microcircuits are the fundamental building blocks of the neocortex [1]. Single instances have been reconstructed experimentally (e.g., [2]), and their general dynamics and information processing capabilities have been investigated theoretically (e.g., [3, 4, 5, 6, 7]). Their architecture is usually represented in connectivity maps consisting of probabilities that neurons establish connections. These maps reduce the complicated circuitry to simple relations between cell types, allowing for efficient instantiations of neural network models in parallel computers [8]. While this approach neglects higher-order features like connectivity motifs, it enables investigations of the links between structural principles of local circuits and their dynamics. Recent years have seen significant advances in the application of electron microscopy (EM) for the reconstruction of local cortical networks, leveraging novel machine learning techniques ([9, 10], but see also [11]). These data allow for a more detailed look into the architecture of local cortical circuits than was previously possible. Here, we construct a layer-resolved, population-based connectivity map from a $1 \:\mathrm{mm}^{3}$ reconstruction of mouse visual cortex [9] with a view towards computational modeling. We compare the obtained microcircuit connectivity based on EM data with an analogous representation from light microscopy (LM) data [2]. In the course of the derivations, we analyze the spatial scale of cortical connectivity. We find that the length scale of this connectivity is consistently overestimated when using morphology-based approaches compared to the actual connectivity available from EM data. The obtained connectivity maps exhibit qualitative similarities, but also stark differences in termination patterns of inter-laminar projections. We finally simulate spiking neural networks constrained by the derived microcircuit architectures with NEST [12]. Systematically varying the input to excitatory and inhibitory populations, we find that in contrast to the model based on LM data, the model based on EM data shows robust asynchronous and irregular firing with a smooth dependence of firing rates on the input parameters, and exhibits a rich dynamical repertoire. We hypothesize that target specificity, namely the increased targeting of inhibitory neurons revealed by the EM data, is an important factor in producing this robustly brain-like spiking activity.