diff --git a/alternator-test/test_projection_expression.py b/alternator-test/test_projection_expression.py
new file mode 100644
index 0000000000..3d640c263f
--- /dev/null
+++ b/alternator-test/test_projection_expression.py
@@ -0,0 +1,184 @@
+# Tests for the various operations (GetItem, Query, Scan) with a
+# ProjectionExpression parameter.
+#
+# ProjectionExpression is an expension of the legacy AttributesToGet
+# parameter. Both parameters request that only a subset of the attributes
+# be fetched for each item, instead of all of them. But while AttributesToGet
+# was limited to top-level attributes, ProjectionExpression can request also
+# nested attributes.
+
+import random
+import string
+import pytest
+import collections
+from botocore.exceptions import ClientError
+
+def random_string(length=10, chars=string.ascii_uppercase + string.digits):
+    return ''.join(random.choice(chars) for x in range(length))
+
+# Utility functions for scan and query into an array of items:
+def full_scan(table, **kwargs):
+    response = table.scan(**kwargs)
+    items = response['Items']
+    while 'LastEvaluatedKey' in response:
+        response = table.scan(ExclusiveStartKey=response['LastEvaluatedKey'], **kwargs)
+        items.extend(response['Items'])
+    return items
+
+def full_query(table, **kwargs):
+    response = table.query(**kwargs)
+    items = response['Items']
+    while 'LastEvaluatedKey' in response:
+        response = table.query(ExclusiveStartKey=response['LastEvaluatedKey'], **kwargs)
+        items.extend(response['Items'])
+    return items
+
+# To compare two lists of items (each is a dict) without regard for order,
+# "==" is not good enough because it will fail if the order is different.
+# The following function, multiset() converts the list into a multiset
+# (set with duplicates) where order doesn't matter, so the multisets can
+# be compared.
+def multiset(items):
+    return collections.Counter([frozenset(item.items()) for item in items])
+
+# Basic test for ProjectionExpression, requesting only top-level attributes.
+# Result should include the selected attributes only - if one wants the key
+# attributes as well, one needs to select them explicitly. When no key
+# attributes are selected, an item may have *none* of the selected
+# attributes, and returned as an empty item.
+@pytest.mark.xfail(reason="ProjectionExpression not yet implemented in GetItem")
+def test_projection_expression_toplevel(test_table):
+    p = random_string()
+    c = random_string()
+    item = {'p': p, 'c': c, 'a': 'hello', 'b': 'hi'}
+    test_table.put_item(Item=item)
+    for wanted in [ ['a'],             # only non-key attribute
+                    ['c', 'a'],        # a key attribute (sort key) and non-key
+                    ['p', 'c'],        # entire key
+                    ['nonexistent']    # Our item doesn't have this
+                   ]:
+        got_item = test_table.get_item(Key={'p': p, 'c': c}, ProjectionExpression=",".join(wanted), ConsistentRead=True)['Item']
+        expected_item = {k: item[k] for k in wanted if k in item}
+        assert expected_item == got_item
+
+# Various simple tests for ProjectionExpression's syntax, using only top-evel
+# attributes.
+@pytest.mark.xfail(reason="ProjectionExpression not yet implemented in GetItem")
+def test_projection_expression_toplevel_syntax(test_table_s):
+    p = random_string()
+    test_table_s.put_item(Item={'p': p, 'a': 'hello', 'b': 'hi'})
+    assert test_table_s.get_item(Key={'p': p}, ConsistentRead=True, ProjectionExpression='a')['Item'] == {'a': 'hello'}
+    assert test_table_s.get_item(Key={'p': p}, ConsistentRead=True, ProjectionExpression='#name', ExpressionAttributeNames={'#name': 'a'})['Item'] == {'a': 'hello'}
+    assert test_table_s.get_item(Key={'p': p}, ConsistentRead=True, ProjectionExpression='a,b')['Item'] == {'a': 'hello', 'b': 'hi'}
+    assert test_table_s.get_item(Key={'p': p}, ConsistentRead=True, ProjectionExpression=' a  ,   b  ')['Item'] == {'a': 'hello', 'b': 'hi'}
+    # It is not allowed to fetch the same top-level attribute twice (or in
+    # general, list two overlapping attributes). We get an error like
+    # "Invalid ProjectionExpression: Two document paths overlap with each
+    # other; must remove or rewrite one of these paths; path one: [a], path
+    # two: [a]".
+    with pytest.raises(ClientError, match='ValidationException'):
+        test_table_s.get_item(Key={'p': p}, ConsistentRead=True, ProjectionExpression='a,a')['Item']
+    # A comma with nothing after it is a syntax error:
+    with pytest.raises(ClientError, match='ValidationException'):
+        test_table_s.get_item(Key={'p': p}, ConsistentRead=True, ProjectionExpression='a,')['Item']
+    with pytest.raises(ClientError, match='ValidationException'):
+        test_table_s.get_item(Key={'p': p}, ConsistentRead=True, ProjectionExpression=',a')['Item']
+    with pytest.raises(ClientError, match='ValidationException'):
+        test_table_s.get_item(Key={'p': p}, ConsistentRead=True, ProjectionExpression='a,,b')['Item']
+    # An empty ProjectionExpression is not allowed. DynamoDB recognizes its
+    # syntax, but then writes: "Invalid ProjectionExpression: The expression
+    # can not be empty".
+    with pytest.raises(ClientError, match='ValidationException'):
+        test_table_s.get_item(Key={'p': p}, ConsistentRead=True, ProjectionExpression='')['Item']
+
+# The following two tests are similar to test_projection_expression_toplevel()
+# which tested the GetItem operation - but these test Scan and Query.
+# Both test ProjectionExpression with only top-level attributes.
+@pytest.mark.xfail(reason="ProjectionExpression not yet implemented in Scan")
+def test_projection_expression_scan(filled_test_table):
+    table, items = filled_test_table
+    for wanted in [ ['another'],       # only non-key attributes (one item doesn't have it!)
+                    ['c', 'another'],  # a key attribute (sort key) and non-key
+                    ['p', 'c'],        # entire key
+                    ['nonexistent']    # none of the items have this attribute!
+                   ]:
+        got_items = full_scan(table,  ProjectionExpression=",".join(wanted))
+        expected_items = [{k: x[k] for k in wanted if k in x} for x in items]
+        assert multiset(expected_items) == multiset(got_items)
+
+@pytest.mark.xfail(reason="ProjectionExpression not yet implemented in Query")
+def test_projection_expression_query(test_table):
+    p = random_string()
+    items = [{'p': p, 'c': str(i), 'a': str(i*10), 'b': str(i*100) } for i in range(10)]
+    with test_table.batch_writer() as batch:
+        for item in items:
+            batch.put_item(item)
+    for wanted in [ ['a'],             # only non-key attributes
+                    ['c', 'a'],        # a key attribute (sort key) and non-key
+                    ['p', 'c'],        # entire key
+                    ['nonexistent']    # none of the items have this attribute!
+                   ]:
+        got_items = full_query(test_table, KeyConditions={'p': {'AttributeValueList': [p], 'ComparisonOperator': 'EQ'}}, ProjectionExpression=",".join(wanted))
+        expected_items = [{k: x[k] for k in wanted if k in x} for x in items]
+        assert multiset(expected_items) == multiset(got_items)
+
+# The previous tests all fetched only top-level attributes. They could all
+# be written using AttributesToGet instead of ProjectionExpression (and,
+# in fact, we do have similar tests with AttributesToGet in other files),
+# but the previous test checked that the alternative syntax works correctly.
+# The following test checks fetching more elaborate attribute paths from
+# nested documents.
+@pytest.mark.xfail(reason="ProjectionExpression does not yet support attribute paths")
+def test_projection_expression_path(test_table_s):
+    p = random_string()
+    test_table_s.put_item(Item={
+        'p': p,
+        'a': {'b': [2, 4, {'x': 'hi', 'y': 'yo'}], 'c': 5},
+        'b': 'hello' 
+        })
+    # Fetching the entire nested document "a" works, of course:
+    assert test_table_s.get_item(Key={'p': p}, ConsistentRead=True, ProjectionExpression='a')['Item'] == {'a': {'b': [2, 4, {'x': 'hi', 'y': 'yo'}], 'c': 5}}
+    # If we fetch a.b, we get only the content of b - but it's still inside
+    # the a dictionary:
+    assert test_table_s.get_item(Key={'p': p}, ConsistentRead=True, ProjectionExpression='a.b')['Item'] == {'a': {'b': [2, 4, {'x': 'hi', 'y': 'yo'}]}}
+    # Similarly, fetching a.b[0] gives us a one-element array in a dictionary.
+    # Note that [0] is the first element of an array.
+    assert test_table_s.get_item(Key={'p': p}, ConsistentRead=True, ProjectionExpression='a.b[0]')['Item'] == {'a': {'b': [2]}}
+    assert test_table_s.get_item(Key={'p': p}, ConsistentRead=True, ProjectionExpression='a.b[2]')['Item'] == {'a': {'b': [{'x': 'hi', 'y': 'yo'}]}}
+    assert test_table_s.get_item(Key={'p': p}, ConsistentRead=True, ProjectionExpression='a.b[2].y')['Item'] == {'a': {'b': [{'y': 'yo'}]}}
+    # Trying to read any sort of non-existant attribute returns an empty item.
+    # This includes a non-existing top-level attribute, an attempt to read
+    # beyond the end of an array or a non-existant member of a dictionary, as
+    # well as paths which begin with a non-existant prefix.
+    assert test_table_s.get_item(Key={'p': p}, ConsistentRead=True, ProjectionExpression='x')['Item'] == {}
+    assert test_table_s.get_item(Key={'p': p}, ConsistentRead=True, ProjectionExpression='a.b[3]')['Item'] == {}
+    assert test_table_s.get_item(Key={'p': p}, ConsistentRead=True, ProjectionExpression='a.x')['Item'] == {}
+    assert test_table_s.get_item(Key={'p': p}, ConsistentRead=True, ProjectionExpression='a.x.y')['Item'] == {}
+    assert test_table_s.get_item(Key={'p': p}, ConsistentRead=True, ProjectionExpression='a.b[3].x')['Item'] == {}
+    # We can read multiple paths - the result are merged into one object
+    # structured the same was as in the original item:
+    assert test_table_s.get_item(Key={'p': p}, ConsistentRead=True, ProjectionExpression='a.b[0],a.b[1]')['Item'] == {'a': {'b': [2, 4]}}
+    assert test_table_s.get_item(Key={'p': p}, ConsistentRead=True, ProjectionExpression='a.b[0],a.c')['Item'] == {'a': {'b': [2], 'c': 5}}
+    # It is not allowed to read the same path multiple times. The error from
+    # DynamoDB looks like: "Invalid ProjectionExpression: Two document paths
+    # overlap with each other; must remove or rewrite one of these paths;
+    # path one: [a, b, [0]], path two: [a, b, [0]]".
+    with pytest.raises(ClientError, match='ValidationException'):
+        test_table_s.get_item(Key={'p': p}, ConsistentRead=True, ProjectionExpression='a.b[0],a.b[0]')['Item']
+    # Two paths are considered to "overlap" if the content of one path
+    # contains the content of the second path. So requesting both "a" and
+    # "a.b[0]" is not allowed.
+    with pytest.raises(ClientError, match='ValidationException'):
+        test_table_s.get_item(Key={'p': p}, ConsistentRead=True, ProjectionExpression='a,a.b[0]')['Item']
+
+# It is not allowed to use both ProjectionExpression and its older cousin,
+# AttributesToGet, together. If trying to do this, DynamoDB produces an error
+# like "Can not use both expression and non-expression parameters in the same
+# request: Non-expression parameters: {AttributesToGet} Expression
+# parameters: {ProjectionExpression}
+@pytest.mark.xfail(reason="ProjectionExpression not yet implemented in GetItem")
+def test_projection_expression_and_attributes_to_get(test_table_s):
+    p = random_string()
+    test_table_s.put_item(Item={'p': p, 'a': 'hello', 'b': 'hi'})
+    with pytest.raises(ClientError, match='ValidationException.*both'):
+        test_table_s.get_item(Key={'p': p}, ConsistentRead=True, ProjectionExpression='a', AttributesToGet=['b'])['Item']