This paper revisits the use of Variational Autoencoders (VAEs) for unsupervised time series anomaly detection (AD) from a frequency perspective. The authors identify challenges in VAE-based methods, particularly in capturing long-periodic heterogeneous patterns and detailed short-periodic trends simultaneously. To address these issues, they propose Frequency-enhanced Conditional Variational Autoencoder (FCVAE), a novel unsupervised AD method for univariate time series. FCVAE integrates both global and local frequency features into the condition of Conditional Variational Autoencoder (CVAE) to enhance the reconstruction accuracy of normal data. Additionally, a "target attention" mechanism is introduced to select the most useful frequency information for better short-periodic trend construction. The proposed FCVAE is evaluated on public datasets and a large-scale cloud system, demonstrating superior performance compared to state-of-the-art methods. The paper also includes comprehensive ablation studies to validate the effectiveness of each component of the FCVAE model.This paper revisits the use of Variational Autoencoders (VAEs) for unsupervised time series anomaly detection (AD) from a frequency perspective. The authors identify challenges in VAE-based methods, particularly in capturing long-periodic heterogeneous patterns and detailed short-periodic trends simultaneously. To address these issues, they propose Frequency-enhanced Conditional Variational Autoencoder (FCVAE), a novel unsupervised AD method for univariate time series. FCVAE integrates both global and local frequency features into the condition of Conditional Variational Autoencoder (CVAE) to enhance the reconstruction accuracy of normal data. Additionally, a "target attention" mechanism is introduced to select the most useful frequency information for better short-periodic trend construction. The proposed FCVAE is evaluated on public datasets and a large-scale cloud system, demonstrating superior performance compared to state-of-the-art methods. The paper also includes comprehensive ablation studies to validate the effectiveness of each component of the FCVAE model.