Parallel programs which run in a shared-memory model have several threads that may interfere. There are constraints between the threads and these constraints can be modelled by a net. We present TLA nets, which are interesting for the representation of concurrent executions. A reduction operation is defined on these nets, in order to detect interferences. These interferences can be eliminated by adding components such as delays to the net. TLA nets are a graphical tool to explore the constraints of parallel programming.