diff --git a/DESCRIPTION b/DESCRIPTION
index 50a9dec98..92dfe06e6 100644
--- a/DESCRIPTION
+++ b/DESCRIPTION
@@ -87,6 +87,7 @@ Collate:
     'backend-postgres-old.R'
     'backend-redshift.R'
     'backend-snowflake.R'
+    'backend-spark-sql.R'
     'backend-sqlite.R'
     'backend-teradata.R'
     'build-sql.R'
diff --git a/NAMESPACE b/NAMESPACE
index a9bff8dd8..17e592bc1 100644
--- a/NAMESPACE
+++ b/NAMESPACE
@@ -39,6 +39,7 @@ S3method(db_col_types,PqConnection)
 S3method(db_col_types,TestConnection)
 S3method(db_col_types,default)
 S3method(db_collect,DBIConnection)
+S3method(db_compute,"Spark SQL")
 S3method(db_compute,DBIConnection)
 S3method(db_connection_describe,DBIConnection)
 S3method(db_connection_describe,MariaDBConnection)
@@ -49,6 +50,7 @@ S3method(db_connection_describe,PostgreSQL)
 S3method(db_connection_describe,PostgreSQLConnection)
 S3method(db_connection_describe,PqConnection)
 S3method(db_connection_describe,SQLiteConnection)
+S3method(db_copy_to,"Spark SQL")
 S3method(db_copy_to,DBIConnection)
 S3method(db_create_index,DBIConnection)
 S3method(db_desc,DBIConnection)
@@ -74,6 +76,7 @@ S3method(dbplyr_as_join_by,default)
 S3method(dbplyr_as_join_by,dplyr_join_by)
 S3method(dbplyr_as_join_by,list)
 S3method(dbplyr_edition,"Microsoft SQL Server")
+S3method(dbplyr_edition,"Spark SQL")
 S3method(dbplyr_edition,ACCESS)
 S3method(dbplyr_edition,HDB)
 S3method(dbplyr_edition,Hive)
@@ -346,6 +349,7 @@ S3method(sql_semi_join,DBIConnection)
 S3method(sql_set_op,DBIConnection)
 S3method(sql_subquery,DBIConnection)
 S3method(sql_table_analyze,"Microsoft SQL Server")
+S3method(sql_table_analyze,"Spark SQL")
 S3method(sql_table_analyze,ACCESS)
 S3method(sql_table_analyze,DBIConnection)
 S3method(sql_table_analyze,HDB)
@@ -361,6 +365,7 @@ S3method(sql_table_analyze,Teradata)
 S3method(sql_table_index,DBIConnection)
 S3method(sql_translate_env,DBIConnection)
 S3method(sql_translation,"Microsoft SQL Server")
+S3method(sql_translation,"Spark SQL")
 S3method(sql_translation,ACCESS)
 S3method(sql_translation,DBIConnection)
 S3method(sql_translation,HDB)
@@ -394,6 +399,7 @@ S3method(sql_values_subquery,Redshift)
 S3method(sql_values_subquery,RedshiftConnection)
 S3method(src_tbls,src_sql)
 S3method(summarise,tbl_lazy)
+S3method(supports_window_clause,"Spark SQL")
 S3method(supports_window_clause,ACCESS)
 S3method(supports_window_clause,DBIConnection)
 S3method(supports_window_clause,Hive)
@@ -500,6 +506,7 @@ export(simulate_oracle)
 export(simulate_postgres)
 export(simulate_redshift)
 export(simulate_snowflake)
+export(simulate_spark_sql)
 export(simulate_sqlite)
 export(simulate_teradata)
 export(sql)
diff --git a/NEWS.md b/NEWS.md
index de9636442..4c2711a84 100644
--- a/NEWS.md
+++ b/NEWS.md
@@ -15,6 +15,8 @@
     (@mgirlich, #1211).
 
   * Joins now work again for Pool and Oracle connections (@mgirlich, #1177, #1181).
+
+* Preliminary databricks Sqark SQL backend (#1377).
   
 * `dbplyr_pivot_wider_spec()` is now exported. Unlike `pivot_wider()` this can
   be lazy. Note that this will be removed soon after `pivot_wider_spec()`
diff --git a/R/backend-spark-sql.R b/R/backend-spark-sql.R
new file mode 100644
index 000000000..79a556107
--- /dev/null
+++ b/R/backend-spark-sql.R
@@ -0,0 +1,130 @@
+#' Backend: Databricks Spark SQL
+#'
+#' @description
+#' See `vignette("translation-function")` and `vignette("translation-verb")` for
+#' details of overall translation technology. Key differences for this backend
+#' are better translation of statistical aggregate functions
+#' (e.g. `var()`, `median()`) and use of temporary views instead of temporary
+#' tables when copying data.
+#'
+#' Use `simulate_spark_sql()` with `lazy_frame()` to see simulated SQL without
+#' converting to live access database.
+#'
+#' @name backend-spark-sql
+#' @aliases NULL
+#' @examples
+#' library(dplyr, warn.conflicts = FALSE)
+#'
+#' lf <- lazy_frame(a = TRUE, b = 1, d = 2, c = "z", con = simulate_spark_sql())
+#'
+#' lf %>% summarise(x = median(d, na.rm = TRUE))
+#' lf %>% summarise(x = var(c, na.rm = TRUE), .by = d)
+#'
+#' lf %>% mutate(x = first(c))
+#' lf %>% mutate(x = first(c), .by = d)
+NULL
+
+#' @export
+#' @rdname backend-spark-sql
+simulate_spark_sql <- function() simulate_dbi("Spark SQL")
+
+#' @export
+`dbplyr_edition.Spark SQL` <- function(con) {
+  2L
+}
+
+#' @export
+`sql_translation.Spark SQL` <- function(con) {
+  sql_variant(
+    base_odbc_scalar,
+    sql_translator(.parent = base_odbc_agg,
+      var = sql_aggregate("VARIANCE", "var"),
+      quantile = sql_quantile("PERCENTILE"),
+      median = sql_aggregate("MEDIAN"),
+      first = function(x, na_rm = FALSE) {
+        check_na_rm(na_rm)
+        glue_sql2(sql_current_con(), "FIRST({.val x})")
+      },
+      last = function(x, na_rm = FALSE) {
+        check_na_rm(na_rm)
+        glue_sql2(sql_current_con(), "LAST({.val x})")
+      },
+    ),
+    sql_translator(.parent = base_odbc_win,
+      var = win_aggregate("VARIANCE"),
+      quantile = sql_quantile("PERCENTILE", window = TRUE),
+      median = win_aggregate("MEDIAN"),
+      first = function(x, order_by = NULL, na_rm = FALSE) {
+        sql_nth(x, 1L, order_by = order_by, na_rm = na_rm, ignore_nulls = "bool")
+      },
+      last = function(x, order_by = NULL, na_rm = FALSE) {
+        sql_nth(x, Inf, order_by = order_by, na_rm = na_rm, ignore_nulls = "bool")
+      },
+      nth = function(x, n, order_by = NULL, na_rm = FALSE) {
+        sql_nth(x, n, order_by = order_by, na_rm = na_rm, ignore_nulls = "bool")
+      },
+    )
+  )
+}
+
+#' @export
+`sql_table_analyze.Spark SQL` <- function(con, table, ...) {
+  # https://docs.databricks.com/en/sql/language-manual/sql-ref-syntax-aux-analyze-table.html
+  glue_sql2(con, "ANALYZE TABLE {.tbl table} COMPUTE STATISTICS")
+}
+
+#' @export
+`supports_window_clause.Spark SQL` <- function(con) {
+  TRUE
+}
+
+#' @export
+`db_copy_to.Spark SQL` <- function(con,
+                                   table,
+                                   values,
+                                   ...,
+                                   overwrite = FALSE,
+                                   types = NULL,
+                                   temporary = TRUE,
+                                   unique_indexes = NULL,
+                                   indexes = NULL,
+                                   analyze = TRUE,
+                                   in_transaction = FALSE) {
+
+  if (!temporary) {
+    cli::cli_abort("Spark SQL only support temporary tables")
+  }
+
+  sql <- sql_values_subquery(con, values, types = types, lvl = 1)
+  db_compute(con, table, sql, overwrite = overwrite)
+}
+
+#' @export
+`db_compute.Spark SQL` <- function(con,
+                                   table,
+                                   sql,
+                                   ...,
+                                   overwrite = FALSE,
+                                   temporary = TRUE,
+                                   unique_indexes = list(),
+                                   indexes = list(),
+                                   analyze = TRUE,
+                                   in_transaction = FALSE) {
+
+  if (!temporary) {
+    cli::cli_abort("Spark SQL only support temporary tables")
+  }
+
+  table <- as_table_ident(table)
+  sql <- glue_sql2(
+    con,
+    "CREATE ", if (overwrite) "OR REPLACE ",
+    "TEMPORARY VIEW {.tbl {table}} AS \n",
+    "{.from {sql}}"
+  )
+  DBI::dbExecute(con, sql)
+
+  table
+}
+
+utils::globalVariables("regexp_replace")
diff --git a/man/backend-spark-sql.Rd b/man/backend-spark-sql.Rd
new file mode 100644
index 000000000..e6b6d93d9
--- /dev/null
+++ b/man/backend-spark-sql.Rd
@@ -0,0 +1,29 @@
+% Generated by roxygen2: do not edit by hand
+% Please edit documentation in R/backend-spark-sql.R
+\name{backend-spark-sql}
+\alias{simulate_spark_sql}
+\title{Backend: Databricks Spark SQL}
+\usage{
+simulate_spark_sql()
+}
+\description{
+See \code{vignette("translation-function")} and \code{vignette("translation-verb")} for
+details of overall translation technology. Key differences for this backend
+are better translation of statistical aggregate functions
+(e.g. \code{var()}, \code{median()}) and use of temporary views instead of temporary
+tables when copying data.
+
+Use \code{simulate_spark_sql()} with \code{lazy_frame()} to see simulated SQL without
+converting to live access database.
+}
+\examples{
+library(dplyr, warn.conflicts = FALSE)
+
+lf <- lazy_frame(a = TRUE, b = 1, d = 2, c = "z", con = simulate_spark_sql())
+
+lf \%>\% summarise(x = median(d, na.rm = TRUE))
+lf \%>\% summarise(x = var(c, na.rm = TRUE), .by = d)
+
+lf \%>\% mutate(x = first(c))
+lf \%>\% mutate(x = first(c), .by = d)
+}