This document describes an approach called Puzzle-based Automatic Testing (PAT) that aims to improve test coverage of automatic test generation techniques by leveraging human problem solving abilities. PAT presents two types of puzzles - object mutation puzzles that require modifying objects to satisfy certain states, and constraint solving puzzles that require finding inputs to satisfy path conditions. The approach was evaluated on two subjects and found to improve test coverage by generating additional test cases from human-solved puzzle solutions. On average, participants were able to solve over 50% of object mutation puzzles and over 70% of constraint solving puzzles within 1 minute each.