This paper addresses the challenge of anomaly detection in spacecraft telemetry, which is crucial for ensuring mission safety and efficiency. Current methods, such as tiered alarms and manual analysis, have limitations, including high false positive rates and the need for extensive expert knowledge. The authors propose a solution that combines Long Short-Term Memory (LSTM) recurrent neural networks (RNNs) with a nonparametric, dynamic thresholding approach to improve accuracy and interpretability. LSTM models are used to predict telemetry values, while the thresholding method evaluates residuals without assuming Gaussian distributions or relying on labeled data. The system is evaluated using real-world, expert-labeled data from the Mars Science Laboratory (MSL) rover and the Soil Moisture Active Passive (SMAP) satellite. Experimental results show that the LSTM-based approach achieves high prediction performance and effective anomaly detection, with a focus on balancing precision and recall. The paper also discusses strategies to mitigate false positives and highlights the importance of command information in anomaly detection. Finally, the authors provide open-source implementations and a dataset for further research.This paper addresses the challenge of anomaly detection in spacecraft telemetry, which is crucial for ensuring mission safety and efficiency. Current methods, such as tiered alarms and manual analysis, have limitations, including high false positive rates and the need for extensive expert knowledge. The authors propose a solution that combines Long Short-Term Memory (LSTM) recurrent neural networks (RNNs) with a nonparametric, dynamic thresholding approach to improve accuracy and interpretability. LSTM models are used to predict telemetry values, while the thresholding method evaluates residuals without assuming Gaussian distributions or relying on labeled data. The system is evaluated using real-world, expert-labeled data from the Mars Science Laboratory (MSL) rover and the Soil Moisture Active Passive (SMAP) satellite. Experimental results show that the LSTM-based approach achieves high prediction performance and effective anomaly detection, with a focus on balancing precision and recall. The paper also discusses strategies to mitigate false positives and highlights the importance of command information in anomaly detection. Finally, the authors provide open-source implementations and a dataset for further research.