Process discovery obtains a process model of activity records. There are two representations of process model, i.e. a probabilistic model and a deterministic model. A deterministic model takes all of activity records to depict a process model, however, the probabilistic model chooses several activity records that satisfy a threshold. Determination of the right threshold leads the emergence of many discovery algorithms of probabilistic models, such as Heuristic Miner, Fodina, Modified Heuristic Miner, and Modified Time-Based Heuristic Miners. Those algorithms determine a threshold based on users or an average of probabilities of activity records, so the quality of the model depends on user proficiency or frequent activities. This paper proposes a new algorithm of probabilistic model discovery, i.e. Heuristic Linear Temporal Logic (HLTL), which determines the threshold based on four quality aspects, i.e. Fitness, Precision, Generalization, and Simplicity. HLTL utilizes Linear Temporal Logic to create a formal representation of process model and store the weight of relationships used for the threshold formation. The result shows that the process model constructing by HLTL has better quality aspects than the process model constructing by Modified Heuristic Miners and Modified Time-Based Heuristic Miners. The generalization value of HLTL is 0.8422 and the generalization value of Modified Heuristic Miner and Modified Time-Based Heuristic Miners are 0.8421.