diff --git a/datasette/templates/table.html b/datasette/templates/table.html
index 65b336b6..1ba61f87 100644
--- a/datasette/templates/table.html
+++ b/datasette/templates/table.html
@@ -104,7 +104,12 @@
     <div class="facet-results">
         {% for facet_info in sorted_facet_results %}
             <div class="facet-info facet-{{ database|to_css_class }}-{{ table|to_css_class }}-{{ facet_info.name|to_css_class }}">
-                <p class="facet-info-name"><strong>{{ facet_info.name }}</strong> <a href="{{ path_with_removed_args(request, {'_facet': facet_info['name']}) }}">&#x2716;</a></p>
+                <p class="facet-info-name">
+                    <strong>{{ facet_info.name }}</strong>
+                    {% if facet_hideable(facet_info.name) %}
+                        <a href="{{ path_with_removed_args(request, {'_facet': facet_info['name']}) }}">&#x2716;</a>
+                    {% endif %}
+                </p>
                 <ul>
                     {% for facet_value in facet_info.results %}
                         <li><a href="{{ facet_value.toggle_url }}">{{ facet_value.label }}</a> {{ "{:,}".format(facet_value.count) }}</li>
diff --git a/datasette/views/table.py b/datasette/views/table.py
index cb5a44f7..ab66b880 100644
--- a/datasette/views/table.py
+++ b/datasette/views/table.py
@@ -537,10 +537,12 @@ class TableView(RowTableShared):
 
         # facets support
         FACET_SIZE = 20
+        metadata_facets = table_metadata.get("facets", [])
+        facets = metadata_facets[:]
         try:
-            facets = request.args["_facet"]
+            facets.extend(request.args["_facet"])
         except KeyError:
-            facets = table_metadata.get("facets", [])
+            pass
         facet_results = {}
         for column in facets:
             facet_sql = """
@@ -719,6 +721,7 @@ class TableView(RowTableShared):
                     key=lambda f: (len(f["results"]), f["name"]),
                     reverse=True
                 ),
+                "facet_hideable": lambda facet: facet not in metadata_facets,
                 "is_sortable": any(c["sortable"] for c in display_columns),
                 "path_with_replaced_args": path_with_replaced_args,
                 "path_with_removed_args": path_with_removed_args,
diff --git a/docs/facets.rst b/docs/facets.rst
index 02ac8d55..51bfc0f5 100644
--- a/docs/facets.rst
+++ b/docs/facets.rst
@@ -3,8 +3,6 @@
 Facets
 ======
 
-This feature is currently under development, see `#255 <https://github.com/simonw/datasette/issues/255>`_
-
 Datasette facets can be used to add a faceted browse interface to any Datasette table. With facets, tables are displayed along with a summary showing the most common values in specified columns. These values can be selected to further filter the table.
 
 Facets can be specified in two ways: using queryset parameters, or in ``metadata.json`` configuration for the table.
@@ -18,43 +16,72 @@ To turn on faceting for specific columns on a Datasette table view, add one or m
 
 This works for both the HTML interface and the ``.json`` view. When enabled, facets will cause a ``facet_results`` block to be added to the JSON output, looking something like this::
 
-    "facet_results": {
-      "state": [
-        {
-          "value": "CA",
-          "count": 10,
-          "selected": false,
-          "toggle_url": "http://...&state=CA"
-        },
-        {
-          "value": "MI",
-          "count": 4,
-          "selected": false,
-          "toggle_url": "http://...&state=MI"
-        }
-      ],
-      "city": [
-        {
-          "value": "San Francisco",
-          "count": 6,
-          "selected": false,
-          "toggle_url": "http://...=San+Francisco"
-        },
-        {
-          "value": "Detroit",
-          "count": 4,
-          "selected": false,
-          "toggle_url": "http://...&city=Detroit"
-        },
-        {
-          "value": "Los Angeles",
-          "count": 4,
-          "selected": false,
-          "toggle_url": "http://...=Los+Angeles"
-        }
-      ]
+    {
+      "state": {
+        "name": "state",
+        "results": [
+          {
+            "value": "CA",
+            "label": "CA",
+            "count": 10,
+            "toggle_url": "http://...?_facet=city_id&_facet=state&state=CA",
+            "selected": false
+          },
+          {
+            "value": "MI",
+            "label": "MI",
+            "count": 4,
+            "toggle_url": "http://...?_facet=city_id&_facet=state&state=MI",
+            "selected": false
+          },
+          {
+            "value": "MC",
+            "label": "MC",
+            "count": 1,
+            "toggle_url": "http://...?_facet=city_id&_facet=state&state=MC",
+            "selected": false
+          }
+        ],
+        "truncated": false
+      }
+      "city_id": {
+        "name": "city_id",
+        "results": [
+          {
+            "value": 1,
+            "label": "San Francisco",
+            "count": 6,
+            "toggle_url": "http://...?_facet=city_id&_facet=state&city_id=1",
+            "selected": false
+          },
+          {
+            "value": 2,
+            "label": "Los Angeles",
+            "count": 4,
+            "toggle_url": "http://...?_facet=city_id&_facet=state&city_id=2",
+            "selected": false
+          },
+          {
+            "value": 3,
+            "label": "Detroit",
+            "count": 4,
+            "toggle_url": "http://...?_facet=city_id&_facet=state&city_id=3",
+            "selected": false
+          },
+          {
+            "value": 4,
+            "label": "Memnonia",
+            "count": 1,
+            "toggle_url": "http://...?_facet=city_id&_facet=state&city_id=4",
+            "selected": false
+          }
+        ],
+        "truncated": false
+      }
     }
 
+If Datasette detects that a column is a foreign key, the ``"label"`` proper will be automatically derived from the detected label column on the referenced table.
+
 Facets in metadata.json
 -----------------------
 
@@ -73,3 +100,19 @@ Here's an example that turns on faceting by default for the ``qLegalStatus`` col
             }
         }
     }
+
+Facets defined in this way will always be shown in the interface and returned in the API, regardless of the ``_facet`` arguments passed to the view.
+
+Suggested facets
+----------------
+
+Datasette's table UI will suggest facets for the user to apply, based on the following criteria:
+
+For the currently filtered data are there any columns which, if applied as a facet...
+
+* Will return 20 or less unique options
+* Will return more than one unique option
+* Will return less unique options than the total number of filtered rows
+* And the query used to evaluate this criteria can be completed in under 20ms
+
+That last point is particularly important: Datasette runs a query for every column that is displayed on a page, which could get expensive - so to avoid slow load times it sets a time limit of just 20ms for each of those queries. This means suggested facets are unlikely to appear for tables with millions of records in them.