The reuse of pilot sequences in a Massive MIMO system leads to pilot contamination, which reduces the channel estimation quality and adds coherent interference in the data transmission. A standard method to reduce pilot contamination, known as regular pilots (RPs), is to increase the pilot overhead and reuse pilots more sparsely in the network. Another approach, denoted as superimposed pilots (SPs), is to send a superposition of pilot and data symbols which allows the system to reuse pilots far more sparsely. This work performs a comparative analysis of RPs and SPs in Massive MIMO considering the joint spectral efficiency (SE) of the uplink (UL) and downlink (DL) communications. A rigorous DL lower bound on the capacity with SPs is derived and multiobjective optimization theory is used to compare the UL and DL SE between RPs and SPs. Numerical results indicate that RPs and SPs give comparable SE when both methods are optimized.