Despite rises in drought frequency and human water demands, streamflow regime shifts from perennial to non‐perennial have not been evaluated in many arid/semi‐arid regions. To document shifts, we created a methodology that classifies streams as naturally perennial or non‐perennial. Our classification used historical, minimally disturbed‐quality USGS streamflow gages (1950–2015) across California. The number of consecutive zero flow days (≥5 days) was used to classify 61% (96/158) and 39% (62/158) of gages as perennial and non‐perennial, respectively. We developed a random forest model to predict flow regime class based on climate and watershed characteristics. To identify regime shifts, we compared the observed class of contemporary (1980–2023) minimally disturbed and disturbed gages with their modeled, natural class. For most minimally disturbed gages, the observed and natural predicted classes were the same, but 13% (7/52) of gages had a modeled perennial regime with an observed non‐perennial class, indicating a drying trend. Among disturbed gages, 22% (64/290) shifted from perennial to non‐perennial and 7% (21/290) from non‐perennial to perennial. Trends in the minimum 7‐day moving average and number of zero‐flow days provided further evidence of drying at minimally disturbed streams, but no pattern at disturbed gages. Our results indicate that few minimally disturbed perennial streams have become non‐perennial to date, but many streams have experienced drying from climate. Streams impacted by human activities had greater drying rates, but regulation has caused some non‐perennial streams to become perennial. By quantifying expected natural streamflow regimes, this work can help monitor, manage, and conserve stream ecosystems.