This work evaluates the performance of several causal structure learning algorithms, in terms of their effectiveness and efficiency in detecting true causal relations among variables. Constraint-based, score-based and hybrid algorithms are jointly compared and ranked according to the two criteria above and their performance is evaluated when used in either directed or undirected acyclic graphs. Fixing the number of variables considered, a Monte Carlo simulation is run for constructing linear causal effects among variables, both in small and large data samples with different causal network properties. Latent confounding variables are empirically demonstrated to be the main drawback of an algorithms’ performance, independently of the size of the sample.