diff --git a/ctis/__init__.py b/ctis/__init__.py
index 8107466..53a6aa8 100644
--- a/ctis/__init__.py
+++ b/ctis/__init__.py
@@ -3,8 +3,10 @@
 spectrograph.
 """
 
+from . import regridding
 from . import scenes
 
 __all__ = [
+    "regridding",
     "scenes",
 ]
diff --git a/ctis/regridding/__init__.py b/ctis/regridding/__init__.py
new file mode 100644
index 0000000..b931a3b
--- /dev/null
+++ b/ctis/regridding/__init__.py
@@ -0,0 +1,9 @@
+"""
+Resampling routines tailored to work with CTIS observations.
+"""
+
+from ._weights import weights
+
+__all__ = [
+    "weights",
+]
diff --git a/ctis/regridding/_weights.py b/ctis/regridding/_weights.py
new file mode 100644
index 0000000..037486f
--- /dev/null
+++ b/ctis/regridding/_weights.py
@@ -0,0 +1,159 @@
+import numpy as np
+import numba
+import named_arrays as na
+
+__all__ = [
+    "weights"
+]
+
+
+def weights(
+    wavelength_input: na.AbstractScalar,
+    position_input: na.AbstractCartesian2dVectorArray,
+    position_output: na.AbstractCartesian2dVectorArray,
+    axis_wavelength_input: str,
+    axis_position_input: tuple[str, str],
+    axis_position_output: tuple[str, str],
+) -> tuple[na.AbstractScalar, dict[str, int], dict[str, int]]:
+    """
+    A version of :func:`named_arrays.regridding.weights` tailored for use with
+    the spatial-spectral cubes in this package.
+
+    Conservatively resampling spatial-spectral cubes is a 3D problem,
+    which can be reduced to a 2D problem :math:`+` a 1D problem since the
+    wavelength coordinates do not vary along the spatial coordinates.
+
+    Parameters
+    ----------
+    wavelength_input
+    position_input
+    position_output
+    axis_wavelength_input
+    axis_position_input
+    axis_position_output
+    """
+
+    wavelength_input = wavelength_input.cell_centers(axis_wavelength_input)
+    position_input = position_input.cell_centers(axis_wavelength_input)
+
+    _weights_2d, shape_input, shape_output = na.regridding.weights(
+        coordinates_input=position_input,
+        coordinates_output=position_output,
+        axis_input=axis_position_input,
+        axis_output=axis_position_output,
+        method="conservative",
+    )
+
+    return
+
+
+
+
+def _weights_3d(
+    weights: na.ScalarArray,
+    shape_inputs: dict[str, int],
+    shape_outputs: dict[str, int],
+    axis_wavelength_input: str,
+    axis_position_input: tuple[str, str],
+    axis_position_output: tuple[str, str],
+) -> tuple[na.AbstractScalar, dict[str, int], dict[str, int]]:
+
+    axes = weights.axes
+
+    result = _weights_3d_numpy(
+        weights=weights.ndarray,
+        axis_wavelength=axes.index(axis_wavelength_input),
+    )
+
+    result = na.ScalarArray(
+        ndarray=result,
+        axes=tuple(ax for ax in axes if ax != axis_wavelength_input),
+    )
+
+    shape_inputs[axis_wavelength_input] = shape_outputs.pop(axis_wavelength_input)
+
+    return result, shape_inputs, shape_outputs
+
+
+def _weights_3d_numpy(
+    weights: np.ndarray,
+    # shape_inputs: tuple[int, ...],
+    # shape_outputs: tuple[int, ...],
+    axis_wavelength: int,
+    # axis_position_input: tuple[int, int],
+    # axis_position_output: tuple[str, str],
+) -> tuple[np.ndarray, tuple[int, ...], tuple[int, ...]]:
+
+    num_wavelength = weights.shape[axis_wavelength]
+
+    weights = np.moveaxis(
+        a=weights,
+        source=axis_wavelength,
+        destination=~0,
+    )
+
+    shape_result = weights.shape[:~0]
+
+    weights = numba.typed.List(weights.reshape(-1))
+
+    weights = _weights_3d_numba(
+        weights=weights,
+        num_wavelength=num_wavelength,
+    )
+
+    weights = np.array(weights).reshape(shape_result)
+
+    return weights
+
+
+@numba.njit()
+def _weights_3d_numba(
+    weights: numba.typed.List,
+    num_wavelength: int,
+) -> numba.typed.List:
+    """
+    Convert a set of 2d weights to 3d.
+
+    Since :mod:`numba` doesn't accept arrays of objects,
+    the 2D weights must be a list of lists of tuples i,j,w.
+
+    It would be nice if the input could be a 3D list of tuples,
+    but constructing this outside of :mod:`numba`-compiled functions is
+    prohibitive.
+    Instead, the user must flatten the first two axes and provide their
+    shapes to reconstruct the 3D inputs.
+
+    Parameters
+    ----------
+    weights
+
+
+    Returns
+    -------
+
+    """
+
+    result = numba.typed.List()
+    result_t = numba.typed.List()
+
+    for d in numba.prange(len(weights)):
+
+        d = numba.types.int64(d)
+
+        w = d % num_wavelength
+
+        weights_tw = weights[d]
+
+        for n in range(len(weights_tw)):
+
+            i_input, i_output, weight = weights_tw[n]
+
+            i_input = (num_wavelength - 1) * i_input + w
+
+            result_t.append((i_input, i_output, weight))
+
+        if w == (num_wavelength - 1):
+            result.append(result_t)
+            result_t = numba.typed.List()
+
+    return result