A comparison of single-cell trajectory inference methods evaluates 45 trajectory inference (TI) methods across 110 real and 229 synthetic datasets to assess their accuracy, scalability, stability, and usability. The study highlights the complementarity of existing TI methods, showing that performance varies depending on dataset characteristics and trajectory topology. The results suggest that the choice of method should be based on the dataset's dimensions and trajectory structure. Guidelines are provided to help users select the best method for their data, and a freely available benchmarking tool (https://benchmark.dynverse.org) is provided for further development of TI methods. TI methods analyze single-cell omics data to infer developmental trajectories, modeling cellular dynamics such as cell cycle, differentiation, and activation. These methods can detect various trajectory topologies, including linear, bifurcating, cyclic, and disconnected graphs. TI methods offer unbiased, transcriptome-wide insights into dynamic processes, enabling the identification of new cell subsets, differentiation trees, and regulatory interactions. Current applications focus on specific cell subsets, but there is a growing need for accurate, scalable, and user-friendly TI methods to analyze increasingly complex single-cell datasets. The study evaluates TI methods based on four criteria: accuracy using gold or silver standard datasets, scalability with respect to cell and feature numbers, stability across dataset subsamples, and usability in terms of software, documentation, and implementation quality. Results show that methods like PAGA, Slingshot, and SCORPIUS perform well across multiple criteria. However, method performance varies significantly across datasets, indicating no single method is universally optimal. The study also highlights challenges in detecting complex topologies and the importance of method development to improve scalability and memory usage. The study finds that most TI methods have poor scalability, with many failing to process large datasets within a reasonable time. Methods with linear time complexity and low memory usage are more scalable. Stability varies among methods, with some producing more consistent results than others. Usability is also assessed, with some methods having better documentation and implementation quality. The study recommends that users try multiple methods to find the best fit for their data, especially when the trajectory topology is unclear. The study concludes that trajectory inference is maturing, particularly for linear and bifurcating trajectories, but challenges remain in accurately inferring complex topologies. Future developments should focus on improving unbiased inference of tree, cyclic, and disconnected topologies, enhancing code assurance and documentation, and designing methods that scale well with increasing cell and feature numbers. The study provides a benchmarking framework and guidelines to help users select and apply TI methods effectively.A comparison of single-cell trajectory inference methods evaluates 45 trajectory inference (TI) methods across 110 real and 229 synthetic datasets to assess their accuracy, scalability, stability, and usability. The study highlights the complementarity of existing TI methods, showing that performance varies depending on dataset characteristics and trajectory topology. The results suggest that the choice of method should be based on the dataset's dimensions and trajectory structure. Guidelines are provided to help users select the best method for their data, and a freely available benchmarking tool (https://benchmark.dynverse.org) is provided for further development of TI methods. TI methods analyze single-cell omics data to infer developmental trajectories, modeling cellular dynamics such as cell cycle, differentiation, and activation. These methods can detect various trajectory topologies, including linear, bifurcating, cyclic, and disconnected graphs. TI methods offer unbiased, transcriptome-wide insights into dynamic processes, enabling the identification of new cell subsets, differentiation trees, and regulatory interactions. Current applications focus on specific cell subsets, but there is a growing need for accurate, scalable, and user-friendly TI methods to analyze increasingly complex single-cell datasets. The study evaluates TI methods based on four criteria: accuracy using gold or silver standard datasets, scalability with respect to cell and feature numbers, stability across dataset subsamples, and usability in terms of software, documentation, and implementation quality. Results show that methods like PAGA, Slingshot, and SCORPIUS perform well across multiple criteria. However, method performance varies significantly across datasets, indicating no single method is universally optimal. The study also highlights challenges in detecting complex topologies and the importance of method development to improve scalability and memory usage. The study finds that most TI methods have poor scalability, with many failing to process large datasets within a reasonable time. Methods with linear time complexity and low memory usage are more scalable. Stability varies among methods, with some producing more consistent results than others. Usability is also assessed, with some methods having better documentation and implementation quality. The study recommends that users try multiple methods to find the best fit for their data, especially when the trajectory topology is unclear. The study concludes that trajectory inference is maturing, particularly for linear and bifurcating trajectories, but challenges remain in accurately inferring complex topologies. Future developments should focus on improving unbiased inference of tree, cyclic, and disconnected topologies, enhancing code assurance and documentation, and designing methods that scale well with increasing cell and feature numbers. The study provides a benchmarking framework and guidelines to help users select and apply TI methods effectively.