How to properly profile Python application performance?

Use cProfile with pstats for broad profiling — it shows function call counts, total time, and cumulative time across your entire stack. For line-level granularity on specific functions, use line_profiler via the @profile decorator. Use memory_profiler if you suspect a leak rather than a CPU bottleneck. Always profile with production-like data because cProfile's overhead is minimal but it can mask contention issues that only appear under real load.