diff --git a/src/components/DimensionsView.tsx b/src/components/DimensionsView.tsx
index 524ca27..1bfa81f 100644
--- a/src/components/DimensionsView.tsx
+++ b/src/components/DimensionsView.tsx
@@ -110,6 +110,16 @@ export function DimensionsView({ data, headers, onEdit, onSaveRow, onCaptureStat
   }, [groups, showOnlyInconsistent]);
 
   const nearDuplicateClusters = useMemo((): NearDuplicateCluster[] => {
+    // Two groups are "similar" if every sorted dimension pair differs by ≤15%
+    // (keys are already sorted ascending, e.g. "15x20x29")
+    const THRESHOLD = 0.15;
+
+    const maxRelativeDiff = (keyA: string, keyB: string): number => {
+      const a = keyA.split('x').map(Number);
+      const b = keyB.split('x').map(Number);
+      return Math.max(...a.map((v, i) => Math.abs(v - b[i]) / Math.max(v, b[i], 0.001)));
+    };
+
     const validGroups = groups.filter(g => g.volume > 0);
     const assignedKeys = new Set<string>();
     const clusters: NearDuplicateCluster[] = [];
@@ -125,13 +135,9 @@ export function DimensionsView({ data, headers, onEdit, onSaveRow, onCaptureStat
         const b = validGroups[j];
         if (assignedKeys.has(b.key)) continue;
 
-        // Check if b is within 5% volume of every group already in the cluster
-        const isNear = cluster.every(g => {
-          const maxVol = Math.max(g.volume, b.volume);
-          return Math.abs(g.volume - b.volume) / maxVol <= 0.05;
-        });
-
-        if (isNear) {
+        // b must be similar to every group already in the cluster
+        const isSimilar = cluster.every(g => maxRelativeDiff(g.key, b.key) <= THRESHOLD);
+        if (isSimilar) {
           cluster.push(b);
           assignedKeys.add(b.key);
         }
@@ -139,9 +145,13 @@ export function DimensionsView({ data, headers, onEdit, onSaveRow, onCaptureStat
 
       if (cluster.length >= 2) {
         const volumes = cluster.map(g => g.volume);
-        const maxVol = Math.max(...volumes);
-        const minVol = Math.min(...volumes);
-        const maxDiffPct = ((maxVol - minVol) / maxVol) * 100;
+        // Max dimension-wise diff across all pairs in the cluster
+        let maxDiffPct = 0;
+        for (let x = 0; x < cluster.length; x++) {
+          for (let y = x + 1; y < cluster.length; y++) {
+            maxDiffPct = Math.max(maxDiffPct, maxRelativeDiff(cluster[x].key, cluster[y].key) * 100);
+          }
+        }
         clusters.push({ groups: cluster, volumes, maxDiffPct });
       }
     }
@@ -244,7 +254,7 @@ export function DimensionsView({ data, headers, onEdit, onSaveRow, onCaptureStat
             <Link2 className="w-4 h-4 text-violet-400" />
             <span className="text-sm font-semibold text-violet-300">Possible data entry errors</span>
             <span className="text-xs text-slate-500 bg-slate-900 px-2 py-0.5 rounded-full border border-slate-700">
-              {nearDuplicateClusters.length} cluster{nearDuplicateClusters.length !== 1 ? 's' : ''} within 5% volume
+              {nearDuplicateClusters.length} cluster{nearDuplicateClusters.length !== 1 ? 's' : ''} with similar dimensions (≤15% per axis)
             </span>
           </div>
           {nearDuplicateClusters.map((cluster, ci) => (
@@ -266,7 +276,7 @@ export function DimensionsView({ data, headers, onEdit, onSaveRow, onCaptureStat
                     </span>
                   ))}
                   <span className="text-xs text-violet-400/70">
-                    — diff. {cluster.maxDiffPct.toFixed(1)}%
+                    — max axis diff {cluster.maxDiffPct.toFixed(1)}%
                   </span>
                 </div>
               </button>