diff --git a/src/eva/transforms/reduce.py b/src/eva/transforms/reduce.py
new file mode 100644
index 00000000..f09f6a94
--- /dev/null
+++ b/src/eva/transforms/reduce.py
@@ -0,0 +1,147 @@
+"""
+EVA Transform: reduce
+=====================
+
+A small, generic reduction transform that operates on a single DataArray in the
+DataCollections and writes the reduced result back into the collection.
+
+1) Fetch a variable from DataCollections (using "Collection::Group::Variable")
+2) Optionally mask fill values to NaN so reductions can use `skipna=True`
+3) Optionally squeeze `binsZDim` by selecting index 0 (common when binsZDim == 1)
+4) Apply a reduction over one or more named dimensions
+5) Save the result back to DataCollections under "new name"
+
+Supported operations: "mean", "sum", "min", "max", "std", and "identity"
+- "identity" does no reduction, but still applies mask/squeeze if requested.
+
+---------------------------------------------------------------------------
+Example YAML
+---------------------------------------------------------------------------
+transforms:
+  - transform: reduce
+    source: "ObsMonitor::griddedBins_ombg_stationPressure::mean"
+    new name: "ObsMonitor::griddedBins_ombg_stationPressure::mean_cycleAvg"
+    op: "mean"
+    dims: ["analysisCycle"]
+    skipna: true
+    squeeze_binsZ: false
+    mask_fill: true
+
+# 2-D output directly (no slicing later):
+  - transform: reduce
+    source: "ObsMonitor::griddedBins_ombg_stationPressure::mean"
+    new name: "ObsMonitor::griddedBins_ombg_stationPressure::mean_cycleAvg_2d"
+    op: "mean"
+    dims: ["analysisCycle"]
+    squeeze_binsZ: true
+    mask_fill: true
+"""
+
+from typing import List, Optional
+
+import xarray as xr
+
+from eva.utilities.logger import Logger
+from eva.utilities.config import get
+from eva.transforms.transform_utils import split_collectiongroupvariable
+
+
+def _mask_fill(da: xr.DataArray) -> xr.DataArray:
+    """Mask the DataArray's _FillValue (if present) to NaN."""
+    if da is None:
+        return da
+    fv = da.attrs.get("_FillValue", None)
+    if fv is None and hasattr(da, "encoding"):
+        fv = da.encoding.get("_FillValue", None)
+    return da.where(da != fv) if fv is not None else da
+
+
+def reduce(config: dict, data_collections) -> None:
+    """
+    Reduce a DataArray over named dimensions and store the result back into EVA.
+
+    Parameters
+    ----------
+    config : dict
+        Transform configuration with the following keys:
+          - source (str): "Collection::Group::Variable" to read
+          - new name (str): "Collection::Group::NewVariable" to write
+          - op (str): one of {"mean","sum","min","max","std","identity"}
+          - dims (list[str], optional): dims to reduce over (e.g., ["analysisCycle"])
+          - skipna (bool, optional, default=True): ignore NaNs in reductions
+          - squeeze_binsZ (bool, optional, default=False):
+                if True and "binsZDim" present, select index 0
+          - mask_fill (bool, optional, default=True):
+                mask `_FillValue` to NaN before reduction
+    data_collections : eva.data.data_collections.DataCollections
+        EVA DataCollections instance.
+
+    Returns
+    -------
+    None
+        The reduced DataArray is added to DataCollections under "new name".
+
+    Notes
+    -----
+    - "identity" op is useful to apply mask/squeeze only (no reduction).
+    - If `dims` is omitted or empty for reduction ops, the transform reduces over *all* dims.
+    - Attributes/coords are preserved by xarray.
+    """
+    logger = Logger("ReduceTransform")
+
+    # Required fields
+    source = get(config, logger, "source")
+    new_name = get(config, logger, "new name")
+    op = get(config, logger, "op").lower()
+
+    # Optional fields
+    dims: Optional[List[str]] = get(config, logger, "dims", default=None)
+    skipna: bool = get(config, logger, "skipna", default=True)
+    squeeze_bins: bool = get(config, logger, "squeeze_binsZ", default=False)
+    mask_fill: bool = get(config, logger, "mask_fill", default=True)
+
+    # Parse source and fetch
+    coll, group, var = split_collectiongroupvariable(logger, source)
+    da: xr.DataArray = data_collections.get_variable_data_array(coll, group, var)
+
+    if da is None:
+        raise ValueError(f"ReduceTransform: source '{source}' not found in DataCollections.")
+
+    # Mask fill value if requested
+    if mask_fill:
+        da = _mask_fill(da)
+
+    # Optional vertical squeeze (select first vertical bin)
+    if squeeze_bins and "binsZDim" in da.dims:
+        da = da.isel(binsZDim=0)
+
+    # Normalize dims: None -> all dims for reduction ops
+    if op != "identity":
+        if dims is None or len(dims) == 0:
+            dims = list(da.dims)
+
+    # Dispatch operation
+    if op == "mean":
+        out = da.mean(dim=dims, skipna=skipna)
+    elif op == "sum":
+        out = da.sum(dim=dims, skipna=skipna)
+    elif op == "min":
+        out = da.min(dim=dims, skipna=skipna)
+    elif op == "max":
+        out = da.max(dim=dims, skipna=skipna)
+    elif op == "std":
+        out = da.std(dim=dims, skipna=skipna)
+    elif op == "identity":
+        out = da
+    else:
+        raise ValueError(f"ReduceTransform: unsupported op '{op}'.")
+
+    # Write back
+    out_coll, out_group, out_var = split_collectiongroupvariable(logger, new_name)
+    data_collections.add_variable_to_collection(out_coll, out_group, out_var, out)
+
+    logger.info(
+        f"ReduceTransform: wrote '{new_name}' "
+        f"(op={op}, dims={dims if op != 'identity' else 'n/a'}, skipna={skipna}, "
+        f"squeeze_binsZ={squeeze_bins}, mask_fill={mask_fill})"
+    )