diff --git a/base/dates/parse.jl b/base/dates/parse.jl
index e638319f10b09..4c2c360c63c0e 100644
--- a/base/dates/parse.jl
+++ b/base/dates/parse.jl
@@ -200,3 +200,58 @@ function Base.parse(::Type{DateTime}, s::AbstractString, df::typeof(ISODateTimeF
     @label error
     throw(ArgumentError("Invalid DateTime string"))
 end
+
+@generated function tryparse_internal{S, F}(::Type{Array{Period}}, str::AbstractString, df::DateFormat{S, F}, raise::Bool=false)
+    token_types = Type[dp <: DatePart ? SLOT_RULE[first(dp.parameters)] : Void for dp in F.parameters]
+    N = length(token_types)
+
+    field_order = sizehint!(Int[], N)
+    field_types = sizehint!(Type[], N)
+    for (i, typ) in enumerate(token_types)
+        if typ <: Period
+            push!(field_order, i)
+            push!(field_types, typ)
+        end
+    end
+    field_parsers = collect(zip(field_order, field_types))
+
+    quote
+        periods = sizehint!(Period[], $(length(field_parsers)))
+        t = df.tokens
+        l = df.locale
+        pos, len = start(str), endof(str)
+
+        err_idx = 1
+        Base.@nexprs $N i->val_i = 0
+        Base.@nexprs $N i->(begin
+            pos > len && @goto done
+            nv, next_pos = tryparsenext(t[i], str, pos, len, l)
+            isnull(nv) && @goto error
+            val_i, pos = unsafe_get(nv), next_pos
+            err_idx += 1
+        end)
+        pos <= len && @goto error
+
+        @label done
+        parts = Base.@ntuple $N val
+        for (i, P) in $field_parsers
+            if i < err_idx
+                push!(periods, P(parts[i]))
+            end
+        end
+        return periods
+
+        @label error
+        # Note: Keeping exception generation in separate function helps with performance
+        raise && throw(gen_exception(t, err_idx, pos))
+        return periods
+    end
+end
+
+function Base.tryparse{T<:Array{Period}}(::Type{T}, str::AbstractString, df::DateFormat)
+    tryparse_internal(T, str, df, false)
+end
+
+function Base.parse{T<:Array{Period}}(::Type{T}, str::AbstractString, df::DateFormat)
+    tryparse_internal(T, str, df, true)
+end
diff --git a/base/deprecated.jl b/base/deprecated.jl
index 9ecfd1843f90f..ad4013e87dd80 100644
--- a/base/deprecated.jl
+++ b/base/deprecated.jl
@@ -1278,6 +1278,14 @@ end
 @deprecate_binding LinearSlow IndexCartesian false
 @deprecate_binding linearindexing IndexStyle false
 
+# #20876
+@eval Base.Dates begin
+    function Base.Dates.parse(x::AbstractString, df::DateFormat)
+        Base.depwarn("`Dates.parse(x::AbstractString, df::DateFormat)` is deprecated, use `sort!(parse(Array{Dates.Period}, x, df), rev=true, lt=Dates.periodisless)` instead.", :parse)
+        sort!(parse(Array{Period}, x, df), rev=true, lt=periodisless)
+     end
+end
+
 # END 0.6 deprecations
 
 # BEGIN 1.0 deprecations
diff --git a/test/dates/io.jl b/test/dates/io.jl
index 7a3790478c2fc..a773ba7635d2e 100644
--- a/test/dates/io.jl
+++ b/test/dates/io.jl
@@ -29,10 +29,18 @@
 # DateTime parsing
 # Useful reference for different locales: http://library.princeton.edu/departments/tsd/katmandu/reference/months.html
 
-let str = "1996/02/15 24:00", format = "yyyy/mm/dd HH:MM"
-    expected = (1996, 2, 15, 24, 0, 0, 0)
-    @test get(Dates.tryparse_internal(DateTime, str, Dates.DateFormat(format))) == expected
+# Using parse(::Array{Period}, ...) allows for more flexibility.
+let str = "02/15/1996 24:00", format = Dates.DateFormat("mm/dd/yyyy HH:MM"), T = Array{Dates.Period}
     @test_throws ArgumentError Dates.DateTime(str, Dates.DateFormat(format))
+
+    expected = [Dates.Month(2), Dates.Day(15), Dates.Year(1996), Dates.Hour(24), Dates.Minute(0)]
+    @test Dates.parse(T, str, Dates.DateFormat(format)) == expected
+end
+
+let T = Array{Dates.Period}
+    expected = Dates.Period[Dates.Year(2017), Dates.Month(3)]
+    @test Dates.parse(T, "2017-03", DateFormat("yyyy-mm-dd")) == expected
+    @test_throws ArgumentError Dates.parse(T, "2017-03-03", DateFormat("yyyy-mm"))
 end
 
 # DateFormat printing